Handbook of Machine Learning for Computational

Optimization: Applications and Case Studies

(Demystifying Technologies for Computational
Excellence) 1st Edition Vishal Jain (Editor)
install download
https://ebookmeta.com/product/handbook-of-machine-learning-for-
computational-optimization-applications-and-case-studies-
demystifying-technologies-for-computational-excellence-1st-
edition-vishal-jain-editor/

Download more ebook from https://ebookmeta.com

We believe these products will be a great fit for you. Click
the link to download now, or visit ebookmeta.com
to discover even more!

Industry 4.0 Technologies for Business Excellence:

Frameworks, Practices, and Applications (Demystifying
Technologies for Computational Excellence) 1st Edition
Shivani Bali (Editor)
https://ebookmeta.com/product/industry-4-0-technologies-for-
business-excellence-frameworks-practices-and-applications-
demystifying-technologies-for-computational-excellence-1st-
edition-shivani-bali-editor/

Artificial Intelligence, Machine Learning, and Data

Science Technologies: Future Impact and Well-Being for
Society 5.0 (Demystifying Technologies for
Computational Excellence) 1st Edition
https://ebookmeta.com/product/artificial-intelligence-machine-
learning-and-data-science-technologies-future-impact-and-well-
being-for-society-5-0-demystifying-technologies-for-
computational-excellence-1st-edition/

Data Science and Innovations for Intelligent Systems:

Computational Excellence and Society 5.0 (Demystifying
Technologies for Computational Excellence) 1st Edition
Kavita Taneja (Editor)
https://ebookmeta.com/product/data-science-and-innovations-for-
intelligent-systems-computational-excellence-and-
society-5-0-demystifying-technologies-for-computational-
excellence-1st-edition-kavita-taneja-editor/

International Human Rights Law and Practice 3rd Edition

Ilias Bantekas

https://ebookmeta.com/product/international-human-rights-law-and-
practice-3rd-edition-ilias-bantekas/
Power and Popular Protest Susan Eckstein (Editor)

https://ebookmeta.com/product/power-and-popular-protest-susan-
eckstein-editor/

Android Programming 1st Edition Bill Phillips Brian

Hardy

https://ebookmeta.com/product/android-programming-1st-edition-
bill-phillips-brian-hardy/

The Gathering Quantum Prophecy 2 Michael Carroll

https://ebookmeta.com/product/the-gathering-quantum-
prophecy-2-michael-carroll/

Medical Internet of Things: Techniques, Practices and

Applications 1st Edition Anirban Mitra

https://ebookmeta.com/product/medical-internet-of-things-
techniques-practices-and-applications-1st-edition-anirban-mitra/

Building a Career in Cybersecurity The Strategy and

Skills You Need to Succeed 1st Edition Yuri Diogenes

https://ebookmeta.com/product/building-a-career-in-cybersecurity-
the-strategy-and-skills-you-need-to-succeed-1st-edition-yuri-
diogenes/
Lives of the eminent philosophers Diogenes Laertius

https://ebookmeta.com/product/lives-of-the-eminent-philosophers-
diogenes-laertius/
 Handbook of Machine
Learning for Computational
Optimization
 Demystifying Technologies for
Computational Excellence: Moving
Towards Society 5.0
Series Editors
Vikram Bali and Vishal Bhatnagar

This series encompasses research work in the feld of Data Science, Edge Computing,
Deep Learning, Distributed Ledger Technology, Extended Reality, Quantum Computing,
Artifcial Intelligence, and various other related areas, such as natural language pro-
cessing and technologies, high-level computer vision, cognitive robotics, automated
reasoning, multivalent systems, symbolic learning theories and practice, knowledge rep-
resentation and the semantic web, intelligent tutoring systems, AI, and education.
The prime reason for developing and growing out this new book series is to focus
on the latest technological advancements – their impact on the society, the challenges
faced in implementation, and the drawbacks or reverse impact on the society due to
technological innovations. With the technological advancements, every individual has
personalized access to all the services, all devices connected with each other commu-
nicating among themselves, thanks to the technology for making our life simpler and
easier. These aspects will help us to overcome the drawbacks of the existing systems
and help in building new systems with latest technologies that will help the society in
various ways, proving Society 5.0 as one of the biggest revolutions in this era.

Industry 4.0, AI, and Data Science

Research Trends and Challenges
Edited by Vikram Bali, Kakoli Banerjee, Narendra Kumar, Sanjay Gour, and
Sunil Kumar Chawla
Handbook of Machine Learning for Computational Optimization
Applications and Case Studies
Edited by Vishal Jain, Sapna Juneja, Abhinav Juneja, and Ramani Kannan
Data Science and Innovations for Intelligent Systems
Computational Excellence and Society 5.0
Edited by Kavita Taneja, Harmunish Taneja, Kuldeep Kumar,
Arvind Selwal, and Ouh Lieh
Artifcial Intelligence, Machine Learning, and Data Science Technologies
Future Impact and Well-Being for Society 5.0
Edited by Neeraj Mohan, Ruchi Singla, Priyanka Kaushal, and Seifedine Kadry

For more information on this series, please visit: https://www.routledge.com/

Demystifying-Technologies-for-Computational-Excellence-Moving-Towards-
Society-5.0/book-series/CRCDTCEMTS
 Handbook of Machine
Learning for Computational
Optimization
Applications and Case Studies

Edited by
Vishal Jain, Sapna Juneja, Abhinav Juneja, and
Ramani Kannan
First edition published 2022
by CRC Press
6000 Broken Sound Parkway NW, Suite 300, Boca Raton, FL 33487-2742

and by CRC Press

2 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN

© 2022 selection and editorial matter, Vishal Jain, Sapna Juneja, Abhinav Juneja, and Ramani
Kannan; individual chapters, the contributors

CRC Press is an imprint of Taylor & Francis Group, LLC

Reasonable eforts have been made to publish reliable data and information, but the author and
publisher cannot assume responsibility for the validity of all materials or the consequences of
their use. Te authors and publishers have attempted to trace the copyright holders of all material
reproduced in this publication and apologize to copyright holders if permission to publish in this
form has not been obtained. If any copyright material has not been acknowledged please write and
let us know so we may rectify in any future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying, microflming, and recording, or in any information
storage or retrieval system, without written permission from the publishers.

For permission to photocopy or use material electronically from this work, access www.copyright.
com or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA
01923, 978-750-8400. For works that are not available on CCC please contact mpkbookspermissions@
tandf.co.uk

Trademark notice: Product or corporate names may be trademarks or registered trademarks and are
used only for identifcation and explanation without intent to infringe.

Library of Congress Cataloging-in-Publication Data

Names: Jain, Vishal, 1983- editor.
Title: Handbook of machine learning for computational optimization :
applications and case studies / Vishal Jain, Sapna Juneja, Abhinav Juneja, Ramani Kannan.
Description: Boca Raton : CRC Press, 2021. | Series: Demystifying
technologies for computational excellence | Includes bibliographical references and index.
Identifers: LCCN 2021017098 (print) | LCCN 2021017099 (ebook) |
ISBN 9780367685423 (hardback) | ISBN 9780367685454 (paperback) |
ISBN 9781003138020 (ebook)
Subjects: LCSH: Machine learning—Industrial applications. | Mathematical
optimization—Data processing. | Artifcial intelligence.
Classifcation: LCC Q325.5 .H295 2021 (print) | LCC Q325.5 (ebook) | DDC 006.3/1—dc23
LC record available at https://lccn.loc.gov/2021017098
LC ebook record available at https://lccn.loc.gov/2021017099

ISBN: 978-0-367-68542-3 (hbk)

ISBN: 978-0-367-68545-4 (pbk)
ISBN: 978-1-003-13802-0 (ebk)

DOI: 10.1201/9781003138020

Typeset in Times
by codeMantra
Contents
Preface...................................................................................................................... vii
Editors....................................................................................................................... xi
Contributors ............................................................................................................xiii

Chapter 1 Random Variables in Machine Learning ............................................. 1

Piratla Srihari

Chapter 2 Analysis of EMG Signals using Extreme Learning Machine

with Nature Inspired Feature Selection Techniques .......................... 27
A. Anitha and A. Bakiya

Chapter 3 Detection of Breast Cancer by Using Various Machine Learning

and Deep Learning Algorithms ..........................................................51
Yogesh Jadhav and Harsh Mathur

Chapter 4 Assessing the Radial Effciency Performance of Bus Transport

Sector Using Data Envelopment Analysis...........................................71
Swati Goyal, Shivi Agarwal, Trilok Mathur, and Nirbhay Mathur

Chapter 5 Weight-Based Codes—A Binary Error Control Coding

Scheme—A Machine Learning Approach......................................... 89
Piratla Srihari

Chapter 6 Massive Data Classifcation of Brain Tumors Using DNN:

Opportunity in Medical Healthcare 4.0 through Sensors .................. 95
Rohit Rastogi, Akshit Rajan Rastogi, D.K. Chaturvedi,
Sheelu Sagar, and Neeti Tandon

Chapter 7 Deep Learning Approach for Traffc Sign Recognition on

Embedded Systems ...........................................................................113
A. Shivankit, Gurminder Kaur, Sapna Juneja, and Abhinav Juneja

v
vi Contents

Chapter 8 Lung Cancer Risk Stratifcation Using ML and AI on Sensor-

Based IoT: An Increasing Technological Trend for Health of
Humanity ...........................................................................................137
Rohit Rastogi, Mukund Rastogi, D.K. Chaturvedi,
Sheelu Sagar, and Neeti Tandon

Chapter 9 Statistical Feedback Evaluation System ............................................153

Alok Kumar and Renu Jain

Chapter 10 Emission of Herbal Woods to Deal with Pollution and Diseases:

Pandemic-Based Threats ...................................................................183
Rohit Rastogi, Mamta Saxena, D. K. Chaturvedi,
and Sheelu Sagar

Chapter 11 Artifcial Neural Networks: A Comprehensive Review................... 203

Neelam Nehra, Pardeep Sangwan, and Divya Kumar

Chapter 12 A Case Study on Machine Learning to Predict the Students’

Result in Higher Education .............................................................. 229
Tejashree U. Sawant and Urmila R. Pol

Chapter 13 Data Analytic Approach for Assessment Status of Awareness of

Tuberculosis in Nigeria..................................................................... 243
Ishola Dada Muraina, Rafeeah Rufai Madaki, and
Aisha Umar Suleiman

Chapter 14 Active Learning from an Imbalanced Dataset: A Study

Conducted on the Depression, Anxiety, and Stress Dataset .............251
Umme Salma M. and Amala Ann K. A.

Chapter 15 Classifcation of the Magnetic Resonance Imaging of the Brain

Tumor Using the Residual Neural Network Framework ...................267
Tina and Sanjay Kumar Dubey

Index...................................................................................................................... 279
Preface
Machine learning is a trusted technology over decades and has fourished on a
global scale touching the lives of each one of us. The modern-day decision making
and processes are all dependent on machine learning technology to make matured
short-term and long-term decisions. Machine learning is blessed to have phenomenal
support from the research community, and have landmark contributions, which is
enabling machine learning to fnd new applications every day. The dependency of
human processes on machine learning-driven systems is encompassing all spheres of
current state-of-the-art systems with the level of reliability it offers. There is a huge
potential in this domain to make the best use of machines in order to ensure the opti-
mal prediction, execution, and decision making. Although machine learning is not a
new feld, it has evolved with ages and the research community round the globe have
made remarkable contribution for the growth and trust of applications to incorporate
it. The predictive and futuristic approach, which is associated with machine learning,
makes it a promising tool for business processes as a sustainable solution. There is
an ample scope in the technology to propose and devise newer algorithms, which are
more effcient and reliable to give machine learning an entirely new dimension in dis-
covering certain latent domains of applications, it may support. This book will look
forward to addressing the issues, which can resolve the modern-day computational
bottom lines which need smarter and optimal machine learning-based intervention
to make processes even more effcient. This book presents innovative and improvised
machine learning techniques which can complement, enrich, and optimize the exist-
ing glossary of machine learning methods. This book also has contributions focusing
on the application-based innovative optimized machine learning solutions, which
will give the readers a vision of how innovation using machine learning may aid in
the optimization of human and business processes.
We have tried to knit this book as a read for all books wherein the learners and
researchers shall get insights about the possible dimensions to explore in their spe-
cifc areas of interest. The chapter-wise description is as follows:
Chapter 1 explores the basic concepts of random variables (single and multiple),
their role and applications in the specifed areas of machine learning.
Chapter 2 demonstrates Wigner-Ville transformation technique to extract the
time-frequency domain features from typical and atypical EMG signals – myopathy
(muscle disorder) and amyotrophic lateral sclerosis (neuro disorder). Nature inspired
feature selection algorithms, whale optimization algorithm (WOA), genetic algo-
rithm (GA), bat algorithm (BA), fre-fy optimization (FA), and particle swarm opti-
mization (PSO) are utilized to determine the relevant features from the constructed
features.
Chapter 3 presents various algorithms of machine learning (ML), which can be
used for breast cancer detection. Since these techniques are commonly used in many
areas, they are also used for making decisions regarding diagnostic diagnosis and
clinical studies.

vii
viii Preface

Chapter 4 measures the effciency and thoroughly explores the scope for opti-
mal utilization of the input resources owned by depots of the RSRTC. The new
slack model (NSM) of DEA is used as it enumerates the slacks for input and output
variables. The model satisfes the radial properties, unit invariance, and translation
invariance. This study enables policy-makers to evaluate inputs for consistent output
up to the optimum level and improve the performance of the ineffcient depots.
Chapter 5 presents a binary error control coding scheme using weight-based
codes. This method is quite used for classifcation and employs the K nearest neigh-
bor algorithms. The paper also discussed the role of distance matrix with hamming
code evaluation.
Chapter 6 exhibits MRI images of the framed brain to create deep neural system
models that can be isolated between different types of heart tumors. To perform this
task, deep learning is used. It is a type of instrument-based learning where the lower
levels responsible for many types of higher-level defnitions appear above the differ-
ent levels of the screen.
Chapter 7 focuses on creating an affordable and effective warning system for driv-
ers that is able to detect the warning sign boards and speed limits in front of the mov-
ing vehicle, and prompt the driver to lower to safer speeds if required. The software
internally works on a deep learning-based modern neural network YOLO (You Only
Look Once) with certain modifcations, which allows it to detect the road signs really
quickly and accurately on low-powered ARM CPUs.
Chapter 8 presents an approach for the classifcation of lung cancer based on the
associated risks (high risk, low risk, high risk). The study was conducted using a lung
cancer classifcation scheme by studying micrographs and classifying them into a
deep neural network using machine learning (ML) framework.
Chapter 9 presents a statistical feedback evaluation system that allows to design an
effective questionnaire using statistical knowledge of the text. In this questionnaire,
questions and their weight are not pre-decided. It is established that questionnaire-
based feedback systems are traditional and quite straightforward, but these systems
are very static and restrictive. The proposed statistical feedback evaluation system
is helpful to the users and manufacturers in fnding the appropriate item as per their
choices.
Chapter 10 presents an experimental work based on the data collected on various
parameters on the scientifc measuring analytical software tools Air Veda instru-
ment and IoT-based sensors capturing the humidity and temperature data from atmo-
spheric air in certain interval of time to know the patterns of pollution increment or
decrement in atmosphere of nearby area.
Chapter 11 concerns with neural network representations and defning suitable
problems for neural network learning. It covers numerous substitute designs for the
primitive units making up an artifcial neural network, such as perceptron units,
sigmoid unit, and linear units. This chapter also covers the learning algorithms for
training single units. Backpropagation algorithm for multilayer perceptron training is
described in detail. Also, the general issues such as the representational capabilities
of ANNs, overftting problems, and substitutes to the backpropagation algorithm are
also explained.
Preface ix

Chapter 12 proposes a system which will make use of the machine learning
approach to predict a student’s performance. Based on student’s current performance
and some measurable past attributes, the end result can be predicted to classify them
among good or bad performers. The predictive models will make students aware who
are likely to struggle during the fnal examinations.
Chapter 13 presents a study that assists in assessing the awareness status of people
on the TB towards its mitigation and serves as contribution to the feld of health infor-
matics. Indeed, the majority of participants claimed that they had low awareness on
the TB and its associated issues in their communities. Though, the participants were
from Kano state, a strategic location in the northern part of Nigeria, which means
that the result of the experiment can represent major opinions of northern residents.
Chapter 14 deals with psychological data related to depression, anxiety, and stress
data to study how the classifcation and analysis is carried out on imbalanced data.
The proposed work not only contributes on providing practical information about
the balancing techniques like SMOTE, but also reveals the strategy for dealing
with working of many existing classifcation algorithms like SVM, Random Forest,
XGBoost etc. on imbalanced dataset.
Chapter 15 proposes the construction of segmented mask of MRI (magnetic reso-
nance image) using CNN approach with the implementation of ResNet framework.
The understanding of ResNet framework using layered approach will provide the
extensive anatomical information of higher-dimensional image for precise clinical
analysis for curative treatment of patients.
Editors
Vishal Jain is an Associate Professor in the Department of CSE at Sharda University,
Greater Noida, India. He has earlier worked with Bharati Vidyapeeth’s Institute of
Computer Applications and Management (BVICAM), New Delhi, India (affliated
with Guru Gobind Singh Indraprastha University, and accredited by the All India
Council for Technical Education). He frst joined BVICAM as an Assistant Professor.
Before that, he has worked for several years at the Guru Premsukh Memorial College
of Engineering, Delhi, India. He has more than 350 research citation indices with
Google scholar (h-index score 9 and i-10 index 9). He has authored more than 70
research papers in reputed conferences and journals, including Web of Science and
Scopus. He has authored and edited more than 10 books with various reputed pub-
lishers, including Springer, Apple Academic Press, Scrivener, Emerald, and IGI-
Global. His research areas include information retrieval, semantic web, ontology
engineering, data mining, ad hoc networks, and sensor networks. He was recipient of
a Young Active Member Award for the year 2012–2013 from the Computer Society
of India, Best Faculty Award for the year 2017, and Best Researcher Award for the
year 2019 from BVICAM, New Delhi.

Sapna Juneja is a Professor in IMS, Ghaziabad, India. Earlier, she has worked as
a Professor in the Department of CSE at IITM Group of Institutions and BMIET,
Sonepat. She has more than 16 years of teaching experience. She completed her
doctorate and masters in Computer Science and Engineering from M.D. University,
Rohtak, in 2018 and 2010, respectively. Her broad area of research is Software
Reliability of Embedded System. Her areas of interest include Software Engineering,
Computer Networks, Operating System, Database Management Systems, and
Artifcial Intelligence etc. She has guided several research theses of UG and PG
students in Computer Science and Engineering. She is editing book on recent tech-
nological developments.

Abhinav Juneja is currently working as a Professor in the Department of IT at KIET

Group of Institutions, Delhi-NCR, Ghaziabad, India. Earlier, he has worked as an
Associate Director and a Professor in the Department of CSE at BMIET, Sonepat.
He has more than 19 years of teaching experience for postgraduate and under-
graduate engineering students. He completed his doctorate in Computer Science
and Engineering from M.D. University, Rohtak, in 2018 and has done masters in
Information Technology from GGSIPU, Delhi. He has research interests in the feld
of Software Reliability, IoT, Machine Learning, and Soft Computing. He has pub-
lished several papers in reputed national and international journals. He has been a
reviewer of several journals of repute and has been in various committees of inter-
national conferences.

xi
xii Editors

Ramani Kannan is currently working as a Senior Lecturer, Center for Smart Grid
Energy Research, Institute of Autonomous system, University Teknologi PETRONAS
(UTP), Malaysia. Dr. Kannan completed Ph.D. (Power Electronics and Drives) from
Anna University, India, in 2012; M.E. (Power Electronics and Drives) from Anna
University, India, in 2006; B.E. (Electronics and Communication) from Bharathiyar
University, India, in 2004. He has more than 15 years of experience in prestigious
educational institutes. Dr. Kannan has published more than 130 papers in various
reputed national and international journals and conferences. He is the editor, co-edi-
tor, guest editor, and reviewer of various books, including Springer Nature, Elsevier
etc. He has received award for best presenter in CENCON 2019, IEEE Conference on
Energy Conversion (CENCON 2019), Indonesia.
Contributors
Shivi Agarwal Renu Jain
Department of Mathematics University Institute of Engineering and
BITS Pilani Technology
Pilani, Rajasthan, India CSJM University
Kanpur, Uttar Pradesh, India
Amala Ann K. A.
Data Science Department Abhinav Juneja
CHRIST (Deemed to be University) KIET Group of Institutions
Bangalore, India Ghaziabad, Uttar Pradesh, India

A. Anitha Sapna Juneja

D.G. Vaishnav College Department of Computer science
Chennai, India IMS Engineering College
Ghaziabad, Uttar Pradesh, India
A. Bakiya
MIT Campus, Anna University Gurminder Kaur
Chennai, India Department of Computer Science and
Engineering
D. K. Chaturvedi BM Institute of Engineering &
Department of Electrical Engineering Technology
DEI, Agra, India Sonepat, India

Sanjay Kumar Dubey Alok Kumar

Department of Computer Science and University Institute of Engineering and
Engineering Technology
Amity University CSJM University
Noida, Uttar Pradesh, India Kanpur, Uttar Pradesh, India

Ayushi Ghosh Divya Kumar

Maulana Abul Kalam Azad University Department of ECE
of Technology IFTMU
Kolkata, West Bengal, India Moradabad, Uttar Pradesh, India

Swati Goyal Rafeeah Rufai Madaki

Department of Mathematics Department of Computer Science
BITS Pilani Yusuf Maitama Sule University
Pilani, Rajasthan, India (Formerly, Northwest University)
Kano, Nigeria
Yogesh Jadhav
Research Scholar Harsh Mathur
Madhyanchal Professional University Department of Computer Science
Bhopal, Madhya Pradesh, India Madhyanchal Professional University
Bhopal, Madhya Pradesh, India

xiii
xiv Contributors

Trilok Mathur Pardeep Sangwan

Department of Mathematics Department of ECE
BITS Pilani MSIT
Pilani, Rajasthan, India Delhi, India

Nirbhay Mathur Tejashree U. Sawant

Department of Electrical & Electronics Department of Computer Science
Universiti Teknologi PETRONAS Shivaji University
Perak, Malaysia Kolhapur, Maharashtra, India

Ishola D. Muraina Mamta Saxena

Department of Computer Science Ministry of Statistics
Yusuf Maitama Sule University Govt. of India
(Formerly, Northwest University) Delhi, India
Kano, Nigeria
A. Shivankit
Neelam Nehra Department of CSE
Department of ECE BM Institute of Engineering &
MSIT Technology
Delhi, India Sonepat, India

Urmila R. Pol Piratla Srihari

Department of Computer Science Department of ECE
Shivaji University Geethanjali College of Engineering and
Kolhapur, Maharashtra, India Technology
Hyderabad, Telangana, India
Akshit Rajan Rastogi
Department of CSE Aisha Umar Suleiman
ABES Engg. College Department of ECE
Ghaziabad, Uttar Pradesh, India Yusuf Maitama Sule University
(Formerly, Northwest University)
Mukund Rastogi Kano, Nigeria
Department of CSE
ABES Engg. College Neeti Tandon
Ghaziabad, Uttar Pradesh, India Research Scholar
Vikram University
Rohit Rastogi Ujjain, Madhya Pradesh, India
Department of CSE
ABES Engg. College Tina
Ghaziabad, Uttar Pradesh, India Department of Computer Science
Engineering
Sheelu Sagar Amity University
Amity International Business School Noida, Uttar Pradesh, India
Noida, Uttar Pradesh, India
Umme Salma M.
Department of Computer Science
CHRIST (Deemed to be University)
Bangalore, India
 1 Random Variables in
Machine Learning
Piratla Srihari
Geethanjali College of Engineering and Technology

CONTENTS
1.1 Introduction ......................................................................................................2
1.2 Random Variable ..............................................................................................3
1.2.1 Defnition and Classifcation.................................................................3
1.2.1.1 Applications in Machine Learning ........................................4
1.2.2 Describing a Random Variable in Terms of Probabilities....................4
1.2.2.1 Ambiguity with Reference to Continuous
Random Variable ...................................................................5
1.2.3 Probability Density Function................................................................6
1.2.3.1 Properties of pdf ....................................................................6
1.2.3.2 Applications in Machine Learning ........................................7
1.3 Various Random Variables Used in Machine Learning...................................7
1.3.1 Continuous Random Variables .............................................................7
1.3.1.1 Uniform Random Variable.....................................................7
1.3.1.2 Gaussian (Normal) Random Variable....................................8
1.3.2 Discrete Random Variables ................................................................ 10
1.3.2.1 Bernoulli Random Variable ................................................. 10
1.3.2.2 Binomial Random Variable ................................................. 11
1.3.2.3 Poisson Random Variable .................................................... 12
1.4 Moments of Random Variable........................................................................ 13
1.4.1 Moments about Origin........................................................................ 13
1.4.1.1 Applications in Machine Learning ...................................... 13
1.4.2 Moments about Mean ......................................................................... 14
1.4.2.1 Applications in Machine Learning ...................................... 14
1.5 Standardized Random Variable...................................................................... 15
1.5.1 Applications in Machine Learning..................................................... 15
1.6 Multiple Random Variables ............................................................................ 16
1.6.1 Joint Random Variables...................................................................... 17
1.6.1.1 Joint Cumulative Distribution Function (Joint CDF)........... 17
1.6.1.2 Joint Probability Density Function (Joint pdf) .................... 17
1.6.1.3 Statistically Independent Random Variables ....................... 18
1.6.1.4 Density of Sum of Independent Random Variables............. 18
1.6.1.5 Central Limit Theorem ........................................................ 19

DOI: 10.1201/9781003138020-1 1
2 Handbook of Machine Learning

1.6.1.6 Joint Moments of Random Variables................................... 19

1.6.1.7 Conditional Probability and Conditional Density
Function of Random Variables ............................................ 22
1.7 Transformation of Random Variables............................................................. 23
1.7.1 Applications in Machine Learning..................................................... 23
1.8 Conclusion ......................................................................................................24
References................................................................................................................24

1.1 INTRODUCTION
Predicting the future using the knowledge about the past is the fundamental objective
of machine learning.
In a digital communication system, a binary data generation scheme referred to as
differential pulse code modulation (DPCM) works on the similar principle, where,
based on the past behaviour of the signal, its future value will be predicted, using a
predictor. A tapped delay line flter serves the purpose. More is the order of the pre-
dictor, better is the prediction, i.e. less is the prediction error.[1]
Thus, machine learning, even though not being referred to by this name earlier,
was/is an integral part of technical world.
The same prediction error with reference to a DPCM system is now being addressed
as confdence interval in connection with machine learning. Less prediction error
implies a better prediction, and as far as machine learning is concerned, the prob-
ability of the predicted value to be within the tolerable limits of error (which is the
confdence interval) should be large, which is a metric for the accuracy of prediction.
The machine learning methodology involves the process of building a statistical
model for a particular task, based on the knowledge of the past data. This collected
past data with reference to a task is referred to as data set.
This way of developing the models to predict about ‘what is going to happen’,
based on the ‘happened’, is predictive modelling.
In detective analysis also, ‘Happened’, i.e. past data, is used, but there is no neces-
sity of predicting about ‘Going to happen’.
For example, 30–35 years back, Reynolds-045 Pen ruled the market for a long
time, specifcally in South India. Presently, the sales are not that much signifcant.
If it is required to study the journey of the pen from past to present, detective anal-
ysis is to be performed, since there is no necessity of predicting its future sales.
Similarly, a study of ‘Why the sales of a particular model of an automobile vehicle
came down?’ also belongs to the same category.
The data set referred above is used by the machine to learn, and hence, it is also
referred to as training data set. After learning, the machine faces the test data. Using
the knowledge the machine gained through learning, it should act on the test data to
resolve the task assigned.
In predictive modelling, if the learning mechanism of the machine is supervised
by somebody, then the mode of learning is referred to as supervised learning. That
supervising ‘somebody’ is the training data set, also referred to as labelled training
data, where each labelled data element such as Di is mapped to a data element D0 .
Such many pairs of elements are the learning resources for the machine, and are used
Random Variables in Machine Learning 3

to build the model, using which the machine predicts, i.e. this knowledge about the
mapped will help the machine to map the test data pairs (input-output pair).
It can be inferred about the supervised learning that there is a target variable
which is to be predicted.
Example: Based on the symptoms of a patient, it is to be predicted whether
he/she is suffering from a particular disease. To enable this prediction, the past
history or statistics such as patients with what symptoms (similar) were categorized
under what disease. This past data (both symptoms and categorization) is the train-
ing data set that supervises the machine in its process of prediction. Here, the target
variable is the disease of the patient, which is to be predicted.
In unsupervised learning mechanism, the training data is considered to be
unlabelled, i.e. only Di. Major functionality of unsupervised learning is pattern
identifcation.
Some of the tasks under unsupervised learning are:

Clustering: Group all the people wearing white (near white) shirts.
Density Estimation: If points are randomly distributed along an axis, the
regions along the axis with minimum/moderate/maximum number points
need to be estimated.

It can be inferred about the unsupervised learning that there is no target variable
which is to be predicted.
With reference to previous example of patient with ill-health, it can be told that
all the people with a particular symptom of ill-health need to be grouped; however,
disease of the patient need not to be predicted, which is the target variable, with refer-
ence to supervised learning.

1.2 RANDOM VARIABLE

1.2.1 DEFINITION AND CLASSIFICATION
For an experiment E to be performed, let S be the set of all possible outcomes of
the experiment (sample space) and ξ be the outcome defned on S. The domain of
X = f (˜ ) is S.
The range of the function depends on the mapping between the outcomes of the
experiment to a numerical value, specifcally real number.
This X is referred to as random variable, and thus, the random variable is a func-
tion defned on S and is real-valued.
Example: For the experiment E = ‘simultaneous throw of two dice’,
S = {(1,1) ,(1,2 ) ,(1,6 ) ,( 2,1) ,(2,2)( 2,6 ) ,( 6,6 )}, where each number of each pair
of this set indicates the face of the particular die.
Each element of S is mapped to a real value by the function

X = f (˜ ) = sum of the two faces

Table 1.1 gives the pairs of all possible outcomes with the corresponding real values
mapped.
4 Handbook of Machine Learning

TABLE 1.1
Sample Space and Mapped Values
Pair in the Sample Space Real Value
(1,1) 2
(1,2), (2,1) 3
(1,3),(2,2),(3,1) 4
(1,4),(2,3),(3,2),(4,1) 5
(1,5),(2,4),(3,3),(4,2),(5,1) 6
(1,6),(2,5),(3,4),(4,3),(5,2),(6,1) 7
(2,6),(3,5),(4,4),(5,3),(6,2) 8
(3,6),(4,5),(5,4),(6,3) 9
(4,6),(5,5),(6,4) 10
(5,6),(6,5) 11
(6,6) 12

This X is referred to as random variable taking all the real values as mentioned.
Thus, a random variable can be considered as a rule by which a real value is
assigned to each outcome of the experiment.
If X = f (˜ ) possesses countably infnite range (points in the range are large in
number, but can be counted), X is referred to as discrete random variable (categorical
with reference to machine learning).
On the other hand, if the range of the function is uncountably infnite (large in
number, which can’t be counted), X is referred to as continuous random variable[2,3]
(similar terminology is used with reference to machine learning also).

1.2.1.1 Applications in Machine Learning

In the process of prediction, which is the major functionality for which the machine
learning is used for, the variables involved are predictor variable and target variable.
The very fundamental task of a machine learning methodology is to identify these
variables with reference to the given task.
A predictor is an independent variable, which is used for prediction, and target
variable is that being predicted and is dependent.
In machine learning terminology, variables are categorical (discrete in nature, e.g.
number of people survived in an accident) and continuous (can have an infnite num-
ber of values between the maximum and the minimum, e.g. age of a person). These
variables are nothing but the discrete and continuous random variables, and are task
specifc. Identifcation of these variables is the primary stage of machine learning.[4]

1.2.2 DESCRIBING A RANDOM VARIABLE IN TERMS OF PROBABILITIES

With reference to the above example, it can be stated that the value mapped (which is
considered as the value taken by the random variable X) is not always 2 or 3 or 4 etc.,
but there is some certainty associated with the mapping, i.e. X will take a value of 7,
7
only under certain outcomes, with a certainty of .
36
Random Variables in Machine Learning 5

Thus, a defnition of a random variable is not complete simply by specifying its

value, without its probabilistic description that deals with the probabilities that X
takes on a specifc value or values.
This description is done by the probability mass function (pmf), which assigns a
probability will for each value of X.
The probability for X = x ( value taken by X ) is P ( X = x ) and will be assigned by
the corresponding pmf.[2]
Example: Consider the case of tossing of an unbiased coin and let this process
of tossing be repeated till a head occurs for the frst time. X = Number of times
of tossing the coin is the random variable. The corresponding sample space can be
concluded as follows:
Since the coin is a fair coin, the event of getting head (H) or tail (T) in each fip is
1
with equal likelihood, i.e. P ( H ) = P ( T ) = .
2
In the frst fip, if a head occurs, there will be no second toss. Then, X = 1 with
1
probability = . On the other hand, if it is a tail, then the user will go to the second
2
fip. If the outcome is a head in the second fip, there will be no third toss. Now,
1
X = −2 with a probability = . This is recurrent.
4
Table 1.2 expresses various values taken by X, with the corresponding probabilities.
The function that assigns the probability for each value taken by X, i.e. the pmf, is
n
˝ 1ˇ
P ( X = n ) = ˆ  , n = 1,2,3
˙ 2˘
The properties possessed by pmf are:

( i ) 0 ˝ P ( X = n ) ˝ 1 ( ii ) ˜P ( X = n) = 1
n

1.2.2.1 Ambiguity with Reference to Continuous Random Variable

Let a discrete random variable X takes values from a set of N values, i.e.
{0, 1, 2,3,… N − 1}, where the values are with equal likelihood. The corresponding
1 1
pmf is P ( X = k ) = , k = 0,1,2,…( N − 1) and lim P ( X = k ) = lim = 0.
N N˝˙ N˝˙ N
Then, the random variable X is considered to be continuous and P ( X = k ) (where
k is a value among N) is found to be zero.
Thus, it can be concluded that the probability of a continuous random variable to
take a specifed value is typically zero.[2]
Under such condition, pmf is not suitable to describe a random variable.

TABLE 1.2
Probability Distribution
xi ( value taken by X ) 1 2 3 4 –
Probability 1 1 1 1 –
2 4 8 16
6 Handbook of Machine Learning

1.2.2.1.1 Cumulative Distribution Function

Instead of X taking a specifc value, consider the case of X to lie in a range,
i.e. X ˜ k , and the corresponding probability P ( X ˛ k ) is referred to as cumulative
distribution function (CDF).[1]
Thus, CDF of a random variable X is FX ( x ) = P ( X ˝ x ), where ‘x’ is the value,
and instead of pmf, this is used to explain a continuous random variable.
Since CDF is also probability, it is bounded as 0 ˛ FX ( x ) ˛ 1.
Properties of CDF are:

• FX ( −˝ ) = 0, since the event X ˜ −˛ will never happen.

• FX ( ˛ ) = 1, since the event X ˜ ° is always true.
• FX ( x1 ) ˛ FX ( x 2 ) , if x1 < x 2, since X ˜ x 2 is a super set of X ˜ x1. Thus,
CDF is a nondecreasing function of X

P ( x1 < X ˝ x 2 ) = FX ( x2 ) − FX ( x1 )

1.2.3 PROBABILITY DENSITY FUNCTION

Even though CDF is a substitute for pmf, to explain a continuous random variable, for
all the random variables, it may not be in a closed form.
ˇ k − m
Example: For a Gaussian random variable Y, the CDF P (Y ˝ k ) = 1 − Q  ,
˘ ˜ 
and the function Q (˜ ) cannot be expressed in a closed form.
Here, m and σ are, respectively, the mean and standard deviation of X.
Under such circumstances, probability density function (pdf) is an alternative tool
to describe a random variable statistically. It is defned for a random variable X at x as

P( x ˙ X < x + ˜ )
f X ( x ) = lim
˜ ˘0 ˜
The certainty or the probability with which X is in the interval ( x , x + ˜ ) is
P( x ° X < x + ˜ ), and the denominator δ is the width (length) of the interval.
Thus, f X ( x ) is the probability normalized by the width of the interval and can be
interpreted as probability divided by width.
From the properties of CDF, P ( x ˝ X < x + ˜ ) = FX ( x + ˜ ) − FX ( x ).
FX ( x + ˜ ) − FX ( x ) d
Hence, f X ( x ) = lim = FX ( x ), i.e. change in CDF is referred
˜ ˇ0 ˜ dx
to as pdf.[2]

1.2.3.1 Properties of pdf

f (x) ˛ 0
ˆ

˜ −ˆ
f ( x ) dx = 1

x
FX ( x ) =
˜−ˇ
f (° ) d °
Random Variables in Machine Learning 7

˜a
f ( x ) dx = P ( a < X < b )

1.2.3.2 Applications in Machine Learning

• Confdence interval is an estimate of a parameter computed based on the
statistical observations of it.
• This specifes a range of reasonable values of being estimated, and the
accuracy of estimation is expressed in terms of confdence level.
• With a given confdence interval, if the number of observations of a
parameter ‘p’ is p1 , p2 , … pn, with the confdence level ɛ, it can be inter-
preted that the estimated value of ‘p’ lies in the given confdence inter-
val with a probability  .
• This is expressed as P(a < X < b) , where X is the parameter being esti-
mated, ( a, b ) is the confdence interval and P(a < X < b) is the conf-
dence level, which can be computed from the probability density of the
parameter being estimated. ˙
• The confdence level P(˜ < X < ° ) can be computed as
˜˝
f ( x ) dx.

• In evaluating a model based on the predicted probabilities (in connec-

tion with binary classifcation), one of the evaluation metrics is area under
curve-receiver-operating characteristic (AUC-ROC) metric.
• This ROC is also referred to as false-positive rate-true-positive rate
curve, which can be obtained from the predicted properties, with refer-
ence to binary classifcation.
• Once this distribution curve is known, area under that curve, i.e.

˜ f ( x ) dx, will be a measure of the accuracy of the predicted.

• Ideally, the area is assumed as 1 (which is the area enclosed by any valid
density function).
• More close to 1 is the area measured, i.e. much better is the perfor-
mance of the predicted model.[4,5]

1.3 VARIOUS RANDOM VARIABLES USED

IN MACHINE LEARNING
1.3.1 CONTINUOUS RANDOM VARIABLES
1.3.1.1 Uniform Random Variable
• The density function of a random variable X (continuous) uniform over
˘ 1
 for ° < x < ˜
(˜ , ° ) is f ( x ) =  ˜ − ° with constant density within
 0 elsewhere

its specifed limits is referred to as uniform random variable (continuous).
• A uniform discrete X takes all the possible values between (˜ , ° ) with
equal likelihood.
8 Handbook of Machine Learning

The physical signifcance of uniform density is that the random variable X can
1
lie in any interval within the limits (˜ , ° ) with the same probability, i.e. ,
˜ −°
for any confdence interval ( k , k + ˜ ) , where ° < k < ˛ , the confidence level
1
P (˜ < X < x + ° ) = .
˛ −˜

• Any confdence level P(˜ 1 < X < °1 ) for the given confdence interval
°1
1
(˜ , ° ) can be obtained as˜˛1 ° − ˛
. dx
• The CDF of uniformly distributed random variable X (continuous) is
 0 for x < ˜

 x −˜
FX ( x ) =  for ˜ < x ˘ °
 ° −˜
 1 for x > °

• A uniform random variable is said to be symmetric about its mean
˙ ˜ + ° ˘ [6]
ˇˆ = .
2 

1.3.1.1.1 Applications in Machine Learning

• When there is no prior realization of the distribution of the variable being
predicted, the variable is considered to lie anywhere in the interval under
consideration with the same probability, i.e. the variable under consider-
ation is treated as uniform.
• In classifcation, decision tree learning is the most widely used algorithm.
The objective of decision tree is to have pure node. The purity of a node and
the information gain are related as:
• More impure is the node, more is the information required to describe.
• More is the information gain, more homogeneous or pure is the node.
• Information gain = 1 − entropy
• More is the entropy, less pure is the node, and vice versa.
• A uniform random variable will have the maximum entropy.
• For example, in a decision tree, in a node if there are two equiprobable
classes, then the corresponding entropy is maximum and is the indica-
tion for more impure node.[5,6]

1.3.1.2 Gaussian (Normal) Random Variable

(
• The density of normally distributed X denoted as N m, ˜ 2 is )
1 ˇ ( x − m )2 
f (x) = exp  − , where ‘m’ and ‘σ ’ are the mean and
2

2˙˜ 2 ˘ 2˜ 2 
variance of X, respectively.
Random Variables in Machine Learning 9

FIGURE 1.1 Gaussian density function.

• This density is a bell-shaped curve, with a point of symmetry at x = m, and

will have its maximum value at x = m (Figure 1.1).
x − m
• The CDF of normal variable X is FX ( x ) = 1 − Q ˘ˆ , where
ˇ ˜ 
1 
ˇ x 2
Q(k ) =
2˙ k˜ exp  −  dx, which is not having a closed form of solution.
˘ 2
• The density curve is symmetrical about its mean, i.e. equal distribution
about its mean.
• If the distribution is more oriented to the right of its mean, it is said to be
right-skewed.
• Similarly, left-skewed distribution can also be identifed.
• Generally, it is preferred to have zero coeffcient of skewness (it is a measure
of symmetry of the given density function).[1]

1.3.1.2.1 Applications in Machine Learning

• Experimental observations in many case studies can be ftted with Gaussian
density.
• Marks of all the students in a class in a particular subject
• Variations in the share price of a particular company
• With reference to machine learning, to make more predictions, more data
is to be added to the model. The following are the ways of adding the data:
• Add external data
• Use existing data more effectively
10 Handbook of Machine Learning

• Feature engineering refers to the technique of generating new features using

existing features. No new data is added.
• Feature pre-processing is one of the primary steps in feature engineer-
ing. This involves updating or transforming the existing features. This is
referred to as feature transformation.
• Feature transformation involves replacing a variable by some mathematical
functions such as ‘log’, ‘square’, ‘square root’, ‘cube’, ‘cube root’ etc.
• If any distribution is right-skewed or left-skewed, it is made normally dis-
tributed using nth root or log and nth power or exponential, respectively.
• As per central limit theorem, the density of sum of ‘n’ number independent
random variables approaches Gaussian density. This is the basis for assum-
ing the channel is normally distributed with reference to a communication
system, which facilitates the study of the noise performance of a communi-
cation system.[7]

1.3.2 DISCRETE RANDOM VARIABLES

1.3.2.1 Bernoulli Random Variable
When an experiment with only two outcomes is repeated independently multiple
number of times, such repetitions are referred to as Bernoulli trials.
Example: The experiment can be

• Single fip of a coin, where the possible outcomes are head and tail
• Identifying about the survival of a person in an accident: survived or not

The pmf of Bernoulli random variable is P ( X = m ) = ( p ) ( q ) , where m assumes

m 1− m

only two values either 0 or 1. It can assume any one value at a time. p and q ( = 1 − p )
are the probability of success and failure, respectively. Success is the required whose
probability is to be computed.
For Example: when an unbiased coin is tossed, if it is required to compute the
1 1−1
˝ 1ˇ ˝ 1ˇ 1
probability of getting a head (represented as m = 1), then P ( X = 1) = ˆ  ˆ  = .
˙ 2˘ ˙ 2˘ 2
Here, success is getting a head.
˘0 for x < 0

The CDF of Bernoulli random variable is FX ( x ) = q for 0 ˙ x < 1[6]

 p + q = 1 for x ˇ 1

1.3.2.1.1 Applications in Machine Learning

• Classifcation is used to predict categorical variables through machine
learning algorithms.
• The test data is to be assigned to a category based on some classifcation
criteria.
Random Variables in Machine Learning 11

• When the variable to be predicted is binary valued, i.e. the test data is to be
categorized into any one of the two classes available, such classifcation is
binary classifcation and the performance of such algorithms can be anal-
ysed using Bernoulli process.[5]

1.3.2.2 Binomial Random Variable

When multiple number of independent Bernoulli trials are repeated, such sequence
of trials is referred to as Bernoulli process. The output of a Bernoulli process is a
binomial random variable/distribution.
Example: Let a fair coin be thrown. Since the outcome can be either head or tail,
it comes under Bernoulli trial. When the experiment is repeated for a number of
times, such sequence of trials is Bernoulli process.
When the experiment is performed ‘r’ times (each experiment being a
Bernoulli trial), the probability of getting the success for ‘k’ times is given as
p ( X = k ) = rck ( p ) ( q ) , where p and q are the probabilities of success and failure,
k r−k

respectively, such that p + q = 1. Here, X is the number of times of having the success
in the experiment.
Here, X is the binomial random variable, since the above probability is the coef-
fcient of kth term in the binomial expansion of ( p + q ) .
r

If it is required to fnd the probability for tail to occur four times, when a fair coin
4 6
˝ 1ˇ ˝ 1ˇ
is tossed ten times, then such probability is p ( X = 4 ) = 10 4 ˆ  ˆ  .
˙ 2˘ ˙ 2˘
The probability of having success for ‘k’ times in ‘r’ number of trials of the exper-
ˆrck ( p ) ( q )
r−k
˝ k
for k = 0,1,2,r
iment in a random order is p ( X = k ) = ˙ , and this
ˆ0
ˇ otherwise
is the pmf of X.
m

The corresponding CDF is FX ( m ) = P ( X ˙ m ) = ˜r

k=0
ck ( p)k ( q )r−k .
Binomial distribution summarizes the number of successes in a series of Bernoulli
experiments, with success probability = p.
Bernoulli distribution is the binomial distribution with single trial.[8]

1.3.2.2.1 Multinoulli Distribution

Similar to Bernoulli distribution used for binary classifcation, multinoulli distribu-
tion also deals with categorical variables. It is a generalization of Bernoulli distribu-
tion (binary classifcation) to a multiclass classifcation, where k is the number of
classes.
Example: In the case of throwing a die, let the sample space be S = {1,2,3,4,5,6}.
Thus, the number of classes can be considered as 6.
Thus, Bernoulli distribution is multinoulli distribution with the number of
classes = 2.[8]
12 Handbook of Machine Learning

1.3.2.2.2 Multinomial Distribution

Multiple number of independent multinoulli trials (e.g. throwing a die multiple num-
ber of times) follows multinomial distribution, which is a generalized binomial dis-
tribution for discrete (categorical) variables/experiments, where each experiment is
with k number of outputs. Here, in n number of trials of an experiment, each experi-
ment has k number of outcomes, which are with the probability of occurrence given
as p1 , p2 ,… pk .[8]

1.3.2.2.3 Applications in Machine Learning

All the algorithms in machine learning may not be able to deal with categorical
variables. In feature pre-processing for categorical variables, to enable this handling,
the categorical variables will be converted to numerical values and this process of
conversion is referred to as variable encoding.
Example: Consider a supermarket having a chain of outlets. Different outlets
in a city are of different size and are graded as small size, medium size, and big
size. A machine learning algorithm cannot deal with such categorical values, i.e.
small, medium and big. In this example, it is the case of multiclass classifcation
with k = 3.
Table 1.3 represents the conversion of the above categorical values into numerical
values.
This process of converting the categorical variables into numeric values is referred
to as one hot encoding and is an example of multinoulli distribution.[6]

1.3.2.3 Poisson Random Variable

When a Bernoulli trial (with a binary outcome, i.e. success or failure) is repeated
independently for multiple number of times (n), the probability of getting the success
(p) for a defned number (m) of times (no restriction on the sequence/order of getting
the success) will be dealt by binomial distribution.
In the limit n ˜ ° and p ˜ 0, i.e. probability of success is infnitesimal, bino-
mial random variable can be approximated as Poisson random variable.
Example: In digital data transmission, when a large number of data bits are being
transmitted, the computation of the probability of bit error will be dealt by this ran-
dom variable.

TABLE 1.3
Variable Encoding
Outlet Size Small Medium Big
1 Big 0 0 1
2 Big 0 0 1
3 Medium 0 1 0
4 Small 1 0 0
5 Medium 0 1 0
Random Variables in Machine Learning 13

Its pmf is p ( X = m ) = e −˜ ( ˜ )m , where ˜ = np is a constant and the corresponding

m!
x
( ° )k .[3]
CDF is FX ( x ) = p ( X ˆ x ) = ˜
k=0
e− °
k!

1.4 MOMENTS OF RANDOM VARIABLE

Moments of a random variable are also referred to as its statistical averages. A ran-
dom variable can have two types of moments: moments about origin and moments
about mean or central moments.

1.4.1 MOMENTS ABOUT ORIGIN

Expected value [E(x)] or mean (m) or average value or expectation of a random value
is referred to as its frst moment about origin and is given as E ( X ) = xi p ( xi ), ˜ i
˙
where p ( xi ) is the certainty with which the random variable X = xi and
˜−˙
f ( x ) dx ,
respectively, for discrete (categorical) and continuous cases.
For a random variable, the second moment about origin is its mean square value
E ˜X
° ˛˝ .
2

Its nth moment E ˜X° ˛˝.

n [1]

1.4.1.1 Applications in Machine Learning

• The averages only are used in the study of random variable, as the certainty
with which takes different values is not unique.
• Linear models are used in regression, when the dependent and independent
random variables are linearly related.
• The preliminary modelling is referred to as benchmark model, where the
mean of the variable will be the solution for the prediction problem.
• Prediction of the relation between the experience of a person and the salary.
• The initial linear modelling will be the benchmark model, taking the
mean, i.e. average of all salaries, i.e. dependent variable as the solution
for the model.
• This may not be the accepted one, since the people with different expe-
riences may have the same salary.
• The model can be improved by introducing the curves of the form
Y = mX + C, i.e. salary = m (Experience) + C, which is a linear model.
• The values of ‘m’ and ‘C’ for which the model gives the best prediction
can be obtained from the cost function, which is given as
n

˜( s − si )
^ 2
i

Cost Function = Mean Square Error (MSE) = i=1

, where si^
n
and si are the predicted and actual ith value, respectively.
14 Handbook of Machine Learning

• This can be referred to as the second moment of the variable Si^ − si .

• Better model results in lower value of MSE.[7,9]

1.4.2 MOMENTS ABOUT MEAN

Let p ( X = 2 ) = p ( X = 6 ) = 0.5, where X is the random variable. Its frst moment
about the origin, i.e. mean, is m = ˜
i
1
2
1
xi p ( xi ) = 2 ˙ + 6 ˙ = 4.
2
To fnd the average amount by which the values taken by X differ from its
mean (the answer is 2), the frst moment about origin or the frst central moment
E [( X − m )] =
i
˜ 1 1
( xi − m ) p( xi ) is defned. But its value ( 2 − 4 ) + ( 6 − 4 ) = 0.
2 2
Thus, the very purpose of defning the frst central moment is not served.
To fnd the same for X, the second central moment E ˆ˙( X − m )2 ˘ˇ = ˜( x − m )
i
i
2
p ( xi )

2 1 2 1
is defned. Its value for the above X is ( 2 − 4 ) + ( 4 − 6 ) = 4. Its positive square
2 2
root is 2, which is the value required.
Thus, E ˝˙( X − m )2 ˆˇ is an indication of the average amount of variation of the val-
ues taken by the random variable with reference to its mean, and hence, its variance
(σ 2), and standard deviation (σ) is its positive square root.
The third central moment of X is E ˝˙( X − m )3 ˆˇ and is referred to as its skew. The
E ˙( X − m )3 ˇ˘
normalized skew, i.e. coeffcient of skewness, is given as ˆ and is a mea-
˜3
sure of symmetry of the density of X.
A random variable with symmetric density about its mean will have zero coef-
fcient of skewness.
If more values taken by the random variable are to the right of its mean, the
corresponding density function is said to be right-skewed and the respective
coeffcient of skewness will be positive (>0). Similarly, the left-skewed density
function can also be specifed and the respective coeffcient of skewness will be
negative (<0).[1]
** Identically distributed random variables will have identical moments.

1.4.2.1 Applications in Machine Learning

• Variance is used to address the concepts of underftting and overftting of a
machine learning model. When the proposed model is trained using differ-
ent data sets that differ signifcantly, at the time of performing on the test
data, the model may not result in the required accuracy. This can be referred
to the error due to variance and is due to the signifcant difference in various
training data models.
• Standard deviation can be used to measure the spread of the data, i.e. it gives
the average distance of a point in a data set from the mean of the data set.
Random Variables in Machine Learning 15

• Learning through decision trees is a widely adopted algorithm in classifca-

tion problems of machine learning.
• In a decision tree, root node represents the entire data.
• Nodes of the tree are divided into sub nodes using the process of
splitting.
• Leaf nodes cannot be further splitted.
• The best split among the available splits is that it results in the most
homogeneous subnodes.
• A higher homogenous node is said to be more purity.
• To decide the best spilt, its variance is used as a metric.
• In this tree, the variance of each child node is computed.
• Weighted average variance of each child node is the variance of the
split.
• The split with less variance is considered to be the best split.
• In feature pre-processing of feature engineering, feature transformation is
applied to a variable through some mathematical functions (as mentioned in
Section 1.3.1.2.1). This transformation is used to make the net distribution
of the variable to be symmetric about its mean, thereby aiming to get zero
coeffcient of skewness.[5,7]

1.5 STANDARDIZED RANDOM VARIABLE

X−m
Let X be a random variable with mean m and variance σ 2. Let X ° = be the new
˜
random variable. Irrespective of the nature of X , X ˜ always will be of zero mean
and unity variance. This X ˜ is referred to as standardized random variable associated
with X.[3]

1.5.1 APPLICATIONS IN MACHINE LEARNING

• Distance-based machine learning algorithms require all the variables (data)
in the same scale.[8,10]
Table 1.4 gives an example where all the data used is not in the same
scale.

TABLE 1.4
Data in Nonuniform Scaling
Loan amount EMI Income
(Rs. in lakhs) (Rs. in thousands) (Rs. in hundreds)
30 50 1600
40 40 1200
50 30 800
16 Handbook of Machine Learning

TABLE 1.5
Statistical Parameters of the Data
Standard deviation (σ) (amount by which value
Variable Mean (m) taken by the variable differs from its mean)
Loan amount 40 10
EMI 40 10
Income 1200 400

TABLE 1.6
Scaled Data
Loan Amount EMI Income
(Rs. in lakhs) (Rs. in thousands) (Rs. in hundreds)
30 − 40 50 − 40 1600 − 1200
= −1 =1 =1
10 10 400
40 − 40 40 − 40 1200 − 1200
=0 =0 =0
10 −10 400
50 − 40 30 − 40 800 − 1200
=1 = −1 = −1
10 −10 400

• All these data are subjected to scaling to make them to be on the same scale,
such that they are comparable. Thus, feature pre-processing requires scal-
ing of the variables.
• Standard scaling is a scaling method, where the scaling of variables is done
X−m
as per the formula X ° = .
˜
• Table 1.5 gives the computations of mean and standard deviation for differ-
ent variables of Table 1.4.
• Table 1.6 represents the above data subjected to scaling.
• It appears that all the scaled variables appear on the same reference scale.
• Thus, the concept of standardized random variable is used in feature
scaling.

1.6 MULTIPLE RANDOM VARIABLES

Data exploration and variable identifcation is one of the constituent stages of machine
learning life cycle.
In data exploration, the process of exploring only one variable at a time and its
study is referred to as univariate analysis.
In bivariate analysis, two variables are studied together. In such contexts, the con-
cept of multiple random variables fnds its place.
Random Variables in Machine Learning 17

1.6.1 JOINT RANDOM VARIABLES

Let E be an experiment and S be the corresponding sample space. Let X and Y be two
variables defned as real functions of S. Then, this pair of variables is referred to as
two-dimensional random variable or joint random variables.

1.6.1.1 Joint Cumulative Distribution Function (Joint CDF)

FXY ( x, y ) = P ( X ˝ x,Y ˝ y ) is the joint CDF of the random variables X and Y, where
x and y are the values taken by them, respectively.[1,11]

1.6.1.1.1 Properties

1. FXY ( −˝, −˝ ) = 0

2. FXY ( ˛, ˛ ) = 1

3. FXY ( −˝, y ) = 0

4. FXY ( x, −˝ ) = 0

5. FXY ( x, ˛ ) = FX ( x ) , which is marginal CDF of X

6. FXY ( ,˛ y ) = FY ( y ) , which is marginal CDF of Y

7. P ( x1 < X ˝ x 2 , y1 < Y ˝ y2 ) = FXY ( x 2 , y2 ) − FXY ( x1 , y2 ) − FXY ( x 2 , y1 )

+ FXY ( x1 , y1 )
8. 0 ˛ FXY ( x, y ) ˛ 1

1.6.1.2 Joint Probability Density Function (Joint pdf)

˝2
The joint pdf of two random variables X and Y is given as f XY ( x , y ) = FXY ( x, y ).
˝x ˙ ˝ y
1.6.1.2.1 Properties

ˆ ˆ
1.
˜ ˜
−ˆ −ˆ
f XY ( x, y ) dx dy = 1

ˆ
2.
˜ y=−ˆ
f XY ( x, y ) dy = f ( x ) , which is the marginal density of X

ˆ
3.
˜ x=−ˆ
f XY ( x, y ) dx = f ( y ) , which is the marginal density of Y

x y
4.
˜ ˜
−ˆ −ˆ
f XY ( x , y ) dx dy = FXY ( x , y )

x2 y2
5. P ( x1 < X < x 2 , y1 < Y < y2 ) =
˜ ˜
x1 y1
f XY ( x , y ) dx dy
18 Handbook of Machine Learning

x ˆ
6. FX ( x ) =
˜ ˜
−ˆ y=−ˆ
f XY ( x , y ) dx dy

ˆ y
7. FY ( y ) =
˜ ˜
x=−ˆ −ˆ
f XY ( x , y ) dx dy [12]

1.6.1.2.2 Joint Occurrence of Random Variables

The joint occurrence of two discrete random variables X and Y can be represented by
a matrix referred to as joint probability matrix P ( X ,Y ), which is given as

XY ˜1 ˜2 … ˜n
°1 p (° 1 , ˜1 ) p (° 1 , ˜ 2 ) … p (° 1 , ˜ n )
P ( X ,Y ) = °2 p (° 2 , ˜1 ) ° … p (° 2 , ˜ n )
° ° ° … °
°n p (° n , ˜1 ) p (° n , ˜ 2 ) … p (° n , ˜ n )

p (˜ i , ° j ) is the joint probability of occurrence of the pair ( X = ˜ , Y = ° j ).[1,3]

1.6.1.2.2.1 Properties of Joint Probability Matrix

1. Each element of the matrix is non-negative

2. ˜˜p(° , ˛ ) = 1
i j
i j

3. ˜p(° , ˛ ) = p( ˛ )
i
i j j

4. ˜p(° , ˛ ) = p(° )
j
i j i

1.6.1.3 Statistically Independent Random Variables

For such X and Y

• FXY ( x, y ) = FX ( x ) FY ( y )

• f XY ( x , y ) = f X ( x ) fY ( y )

• P ( x1 < X < x 2 , y1 < Y < y2 ) = P ( x1 < X < x 2 ) P ( y1 < Y < y2 )

1.6.1.4 Density of Sum of Independent Random Variables

Let Z = X + Y , where X and Y are the random variables with individual den-
sity functions f X ( x ) and fY ( y ), respectively, and are independent.[3] Then,
Random Variables in Machine Learning 19

ˆ ˆ
f Z ( z ) = f X ( x ) * fY ( y ) =
˜ −ˆ
f X ( x ) fY ( z − x ) dx =
˜
−ˆ
f X ( z − y ) fY ( y ) dy, which is
convolution of their individual density functions.
This principle can be extended to multiple number of independent random vari-
ables also.

1.6.1.5 Central Limit Theorem

This theorem deals with the density of sum of independent random variables.
Central limit theorem can be stated as ‘density of sum of n number of independent
identically distributed random variables approaches Gaussian density, in the limit
n ˜ °’.
This is true in the case of distinct distributions also.[3]

1.6.1.6 Joint Moments of Random Variables

1.6.1.6.1 Joint Moments about Origin

For two random variables X and Y , mnk = E  X nY k  is the ( n + k ) th-order joint
moment o about origin.
For discrete X and Y , mnk = ∑∑
xin y kj ⋅ p ( xi , y j ) and for continuous
i j
∞ ∞
X and Y , mnk =
∫ ∫
−∞ −∞
x y f ( x , y ) dx dy.
n k

For two random variables X and Y, there will be three number of second-order
joint moments.
They are

m20 = E ˝˙ X 2 ˆˇ = Mean Square value of X m02 = E ˝˙Y 2 ˆˇ = Mean Square value of Y

m11 = E [ XY ] = RXY , Correlation between X and Y

For two independent random variables, E °̃ X nY k ˛˝ = E °̃ X n ˛˝ E °̃Y k ˛˝.

If the correlation RXY = 0, X and Y are said to be orthogonal.
Independently, X and Y with zero individual means are orthogonal.

1.6.1.6.2 Joint Central Moments

mnk = E ˙( X − mx ) (Y − m y ) ˇ is referred to as ( n + k ) th-order joint central moments
n k
ˆ ˘
or joint moment about mean of two random variables X and Y.
For two random variables X and Y, there will be three number of second-order
joint moments. They are

m20 = E ˆˇ( X − mx ) ˘ = ˜ x2 , variance of X m02 = E ˆ(Y − m y ) ˘ = ˜ y2 , variance of Y

2 2
ˇ 

m11 = E ˇˆ( X − mx )(Y − m y ) ˘ = ˜ XY , covariance of X and Y , Cov ( X ,Y ) ,

which is zero for two independent random variables.

20 Handbook of Machine Learning

Let X ′ and Y ′ be two standardized variables associated with X and Y.

The second-order joint central moment m11 of X ′ and Y ′ is referred to as correla-
tion coefficient ( ρ XY ) of X and Y.

 X − mx    Y − m y    E ( X − mx )(Y − m y )  σ
ρ XY = E [ X ′Y ′ ] = E       = = XY
 σ x    σ y    σ xσ y σ xσ y

It is an indication of similarity between random variables and is bounded as

−1 ≤ ρ ≤ 1.
If the X = + βY , ( β being a constant ), irrespective of the value of β , ρ XY = +1,
which indicates that X and Y are similar.
If X = − βY , irrespective of the value of β, ρ XY = −1, which indicates that X and
Y are dissimilar.
A value of ˜ XY = +1 indicates the maximum similarity between X and Y, while
˜ XY = −1 is an indication of maximum dissimilarity.
Correlation coeffcient is used to quantify the relation between two random
variables.
A value of ˜ XY = 0 is an indication of the uncorrelated nature of X and Y.
For two jointly Gaussian random variables, independence implies uncorrelated
nature, and vice versa.[3]

1.6.1.6.3 Applications in Machine Learning

• Covariance is the simultaneous variation for two random variables.
• Linear regression involves only one predictor, where a linear relation-
ship is assumed between the dependent variable and the predictor vari-
able (independent).
• In linear regression models, slope-intercept format Y = p + qX is used.
• Here, X , Y , p, and q are the independent (predictor) variable, depen-
dent (being predicted) variable, intercept and the slope of the linear
model, respectively.
• The test data regarding X and Y is fed to the model, and the correspond-
ing p and q are to be identifed that relate X and Y.
• The error in identifying the values of p and q is the marginal error or
residual error.
• This modifes the slope-intercept format as Y = p + qX + ˜ .
• Ordinary least-squares technique is used to estimate the straight line
(linear model) that minimizes the difference between the actual and
predicted values of Y, which is the error.
• For each value of the predicted, this error can be calculated and the sum
of the squares of these errors (SSE) is
Cov ( X ,Y ) i
˜
° i2 , which is found to be least
for q = .
Var ( X )
• From this value of q, the corresponding value of p can be computed.
• The linear relationship and direction of that relation between two random
variables can be measured using the correlation coeffcient.
Random Variables in Machine Learning 21

• In a passenger ticketing system, the variables can be the age of a person

and the ticket fare.
• To develop a model for the relation between these two variables, their
correlation coeffcient is a metric used, with age being the independent
and fare of the ticket being the dependent variable.
• The possibilities in the linear modelling can be

Fare = K ( Age ) ˝ fare of the ticket increases with age

Fare = −K ( Age ) ˙ fare of the ticket decreases with age

– Fare vs. age variation is constant → fare is independent of age

• Similar inference can be made from the correlation coeffcient between
these variables (Figure 1.2).

FIGURE 1.2 Correlation coeffcient between the variables.

22 Handbook of Machine Learning

• Thus, correlation coeffcient can be used as the metric for the measure
of the linear relation between the variables.
• Variance and standard deviation are the measures of the spread of the data
set around its mean and are one-dimensional measure. When dealing with
data sets with two dimensions, the relation between these two variables
in the data set (e.g. number of hours spent by a tailor in stitching and the
number of shirts stitched can be the variables) i.e. the statistical analysis
between the two variables will be studied using covariance. Covariance for
one random variable is nothing but its variance. In the case of n variables,
covariance matrix [ n X n ] is used for the statistical analysis of all the pos-
sible pairs of the variables.[9,13,14]

1.6.1.7 Conditional Probability and Conditional

Density Function of Random Variables
During the study of bivariate random variables, conditional probability is also of
good importance.
This deals with the conditional occurrence of an event, while the occurrence of

( )
other event being the condition is denoted as p A B =
( ˜ B) = p( A, B) , which
p A
P ( B) p( B)
is the probability of event A, under the occurrence of the event B.

For categorical variables X ( taking values x1 , x 2 ,…x n ) and Y ( taking values

p ( X = xi , Y = y j )
(
y1 , y2 ,…yn ), the probability p X = xi Y = y =
j ) p (Y = y j )
.
On similar lines, for continuous X and Y, density function is used in the place of
probabilities.

1.6.1.7.1 Properties of Conditional Density Function

( )
1. f X x y is always non-negative
Y

( )
ˆ
2.
˜
−ˆ
f X x y dx = 1
Y

Similar properties hold good for discrete variables also, but defned under discrete
summation.[15]

1.6.1.7.2 Applications in Machine Learning

( )
p B A p( A)
( )
• Baye’s theorem is stated as p A B =
p( B)
.
• Let the data set consist of various symptoms leading to corona/malaria/
typhoid.
Random Variables in Machine Learning 23

• The prior knowledge about the certainty (probability) of such various

hypotheses is referred to as prior, and such probability is a priori probability.
• Based on the data set, the conclusion about the patient is to be made.
• Using above such hypothesis, based on the symptoms of the patient, conclu-
sions are to be drawn whether the patient is suffering from corona/malaria/
typhoid etc., i.e. which of these hypotheses is applicable for the patient and
with what probability. Such probability is a posteriori probability.

( )
• Then, the probability p having a specific disease Symptom S , i.e. having a
specifc symptom S, the probability of suffering from a specifc disease is
the conditional probability.
• If all the hypotheses are of equal a priori probability, then the above condi-
tional probability can be obtained from the probability of having those symp-
(
toms, knowing the disease, i.e. p Symptom S having the specific disease . )
This probability is referred to as maximum likelihood (ML) of the specifc
hypothesis.[7,9]
• Then, the required conditional probability is

(
p having a specific disease Symptom S )
=
( )
p Symptom S having the specific disease ˝ p ( having the specific disease )
p ( Symptom S )

1.7 TRANSFORMATION OF RANDOM VARIABLES

Any system, depending on its performance, transforms the input variable X into
the output variable Y, with the transformation being the system’s performance, i.e.
Y = g(X), where g() is the transformation.
This transformation can be (1) monotonic and (2) non-monotonic.
dx
The unknown density of the resulting variable Y is given as f ( y ) = f ( x ) | x = x1 ˝ 1 ,
dy
where x1 is the real solution of x from the transformation.
If the transformation results in multiple number of real solutions for X,

f ( y) =
i
˜ dx
f ( x ) | x = xi . i [3]
dy

1.7.1 APPLICATIONS IN MACHINE LEARNING

• Regression models are used to identify a functional relation between the
dependent (target) variable and the predictor variable.
• Based on this relation, future values of the target variable are to be pre-
dicted as a function of predictor variable.
• In linear regression, this functional relation is assumed to be linear and of
the form ˜ = p + q° .
24 Handbook of Machine Learning

TABLE 1.7
Linear Regression Transformation of Variables[10]
Nonlinear
Relations Reduced to Linear Law
˜ = p° n log ( ˜ ) = log ( p ) + n ˇ log (° ) ˘ Y = nX + C , with Y = log ( ˜ ) , X = log (° ) ,C = log ( x )
˜ = m° n + C Y = mX + C, with X = ˜ n , Y = °
˜ = p° n + q.log (° ) ˜ °n
Y = aX + b, with Y = ,X = , a = p, b = q
log (° ) log (° )
˜ = pe q˛ Y = m˜ + c, with Y = log ( ° ) , m = q ˇ log ( e ) ,
c = log ( p )

• If the existing relationship is not linear, the dependent variable is subjected

to some transformations to convert the existing nonlinear relation to linear.[8]
• Table 1.7 specifes some of the transformations.

1.8 CONCLUSION
Thus, random variables are playing a vital role in the felds of machine learning and
artifcial intelligence. Since prediction about the future values of a variable involves
some amount of uncertainty, theory of probability and random variables are essen-
tial constituent building blocks of the algorithms used to teach a machine to perform
certain tasks that are dealing with the principles of learning based on the experience.
These random variables are very much specifc in the theory of signal estimation too.[11]

REFERENCES
1. Bhagwandas P. Lathi and Zhi Ding – Modern Digital and Analog Communication
Systems, Oxford University Press, New York, International Fourth Edition, 2010.
2. Scott L. Miller and Donald G. Childers – Probability and Random Processes with
Applications to Signal Processing and Communications, Academic Press, Elsevier
Inc., Boston, MA, 2004.
3. Henry Stark and John W. Woods – Probability and Random Processes with Applications
to Signal Processing, Pearson, Upper Saddle River, NJ, Third Edition, 2002.
4. Kevin P. Murphy – Machine Learning: A Probabilistic Perspective, The MIT Press,
Cambridge, MA, 2012.
5. Jose Unpingco – Python for Probability, Statistics, and Machine Learning, Springer,
Cham, 2016.
6. Steven M. Kay – Intuitive Probability and Random Processes using MATLAB, Springer,
New York, 2006.
7. Peter D. Hoff – A First Course in Bayesian Statistical Methods, Springer, New York,
2009.
8. Shai Shalev-Shwartz and Shai Ben-David – Understanding Machine Learning: From
Theory to Algorithms, Cambridge University Press, New York, 2014.
9. Bernard C. Levy – Principles of Signal Detection and Parameter Estimation, Springer,
Cham, 2008.
Random Variables in Machine Learning 25

10. Michael Paluszek and Stephanie Thomas – MATLAB Machine Learning, Apress,
New York, 2017.
11. Rober M. Gray and Lee D. Davisson – An Introduction to Statistical Signal Processing,
Cambridge University Press, Cambridge, 2004.
12. Friedrich Liese and Klaus-J. Miescke – Statistical Decision Theory – Estimation,
Testing and Selection-Springer Series in Statisitcs, Springer, New York, 2008.
13. James O. Berger – Statistical Decision Theory and Bayesian Analysis, Springer-Verlag,
New York Inc., New York, Second Edition, 2013.
14. Ruise He and Zhiguo Ding (Eds.) – Applications of Machine Learning in Wireless com-
munications, IET Telecommunication Series 81, IET The Institution of Engineering
and Technology, London, 2019.
15. Robert M. Fano – Transmission of Information: A Statistical Theory of Communications,
The MIT Press, Cambridge, MA, 1961.
 2 Analysis of EMG
Signals using Extreme
Learning Machine with
Nature Inspired Feature
Selection Techniques
A. Anitha
D.G. Vaishnav College

A. Bakiya
MIT Campus, Anna University

CONTENTS
2.1 Introduction..................................................................................................... 27
2.2 Data Set........................................................................................................... 30
2.3 Feature Extraction........................................................................................... 30
2.4 Nature Inspired Feature Selection Methods.................................................... 32
2.4.1 Particle Swarm Optimization Algorithm (PSO) ................................. 32
2.4.2 Genetic Algorithm (GA) ..................................................................... 33
2.4.3 Fire-Fly Optimization Algorithm (FA) ...............................................34
2.4.4 Bat Algorithm (BA) ............................................................................ 36
2.4.5 Whale Optimization Algorithm (WOA) ............................................. 37
2.4.5.1 Exploitation Phase ............................................................... 37
2.4.5.2 Exploration Phase ................................................................ 38
2.5 Extreme Learning Machine (ELM) ................................................................ 39
2.6 Results and Discussion ................................................................................... 41
2.7 Conclusion ...................................................................................................... 45
References................................................................................................................ 47

2.1 INTRODUCTION
Neuromuscular Impairment (NMI) is an ailment that affects the neuromuscular
system by breaking in the communication path between the muscles and the ner-
vous system (Reed et al. 2017). The symptoms of NMI include muscle numbness,
fatigue muscles, abnormal pain sensation, atrophy in muscles, and fasciculation in

DOI: 10.1201/9781003138020-2 27
28 Handbook of Machine Learning

muscles. Amyotrophic Lateral Sclerosis (ALS) is a progressive degenerative NMI

and more dreadful condition, which becomes severe when it is not treated properly
(Blottner & Salanova 2015). Myopathy is a NMI, which majorly affects the mus-
cles. Electromyogram (EMG) is a galvanic activity used by medical practitioners for
recording the muscle activity, which assists in identifying the NMI (Sadikoglu et al.
2017). EMG helps in indicating the symptoms of muscles and nerves. However, to
identify the abnormalities, timely intervention is required to prevent further deterio-
ration of neuromuscular system.
Various computer-assisted methods are available in the literature (Tuncer et al.
2020; Khan et al. 2019; Too et al. 2017, Subasi et al. 2018) in identifying normal and
NMI impairments from the signals. However, this chapter focuses on identifying and
analyzing the abnormalities that exist in EMG (ALS and myopathy) signals using the
Extreme Learning machine (ELM) with Nature Inspired Feature Selection (NIFS)
techniques.
Feature Construction (FC) is a crucial process in extracting the original and novel
features from the EMG signals. FC elevates the prediction accuracy of the classifer by
deriving features from the given input data (Phinyomark et al. 2012). FC techniques
also reduce dimensionality reduction combining the original features to form a novel
feature. FC techniques for signals are accumulated as time-domain (TD) (Phinyomark
et al. 2012), frequency-domain (FD) (Subasi et al. 2018), and time-frequency (TF)
(Subasi et al. 2018; Phinyomark et al. 2012) methods. Extraction of TD features uses
mathematical properties and does not involve any transformation. In contrast, extrac-
tion of frequency features requires Fourier transform. However, both time and fre-
quency features assume stationary signals for precise results (Zawawi et al. 2018).
The combined features of TD and FD are referred as time-frequency (TF) features.
The TF features provide highly nonstationary information of signals (Oskoei & Hu
2007). Extraction of TF features is more appropriate, since EMG signals are nonsta-
tionary signals. However, the TF transformation techniques are mandatory to con-
vert single-dimensional time series of EMG signals into two-dimensional TF images.
Many transformation techniques are available to transform the EMG signals into TF
features (Bakiya et al. 2020; Ambikapathy et al. 2018). In this chapter, Wigner-Ville
transformation is employed in transforming the TF features.
In machine learning applications, real-life problems demand enormous number of
features. From huge number of features, identifying essential features after eliminat-
ing redundant and irrelevant features is crucial for machine learning task. Feature
Selection (FS) is an indispensable task in enhancing the capabilities of machine
learning and data mining problems by reducing data dimensionality, computational
time, and assists in constructing simplifed learning model (Kira & Rendell 1992).
FS techniques focus on selecting feature subset from the original feature space with-
out any transformations. In contrast, feature extraction constructs the features from
the dataset. The best feature subset accommodates reduced dimensions, concurrently
enhancing the degree of accuracy by discarding the irrelevant features. In addition,
the complexity of FS techniques relies on the extent of search space. Figure 2.1
depicts the essential steps of FS process, which includes the generation of feature
subset, evaluation of feature subset, termination criterion and validation of perfor-
mance (Dash & Liu 1997).
Analysis of EMG Signals using ELM 29

FIGURE 2.1 Feature selection process.

The subset evaluation is a pivotal step in estimating the effectiveness of the feature
subset using evaluation criteria. Examining the evaluation of features, FS techniques
can be categorized as flter methods (Guyon & Elisseeff, 2003), wrapper methods
(Chandrashekar & Sahin 2014), and embedded methods (Stańczyk 2015). Filter
model depends on the common data characteristics, and learning model is not used
for evaluating feature subsets. In flter model, features are ranked, and it is evaluated
independently (univariate) or in batch (multivariate). The chosen feature subsets con-
tain features with highest scores. The features with highest score are grouped as fea-
ture subset. Filter model takes an advantage of less computational complexity (Duch
2006); alternatively during classifcation task, it excludes the performance of selected
features. Wrapper model utilizes a learning algorithm (Chandrashekar & Sahin 2014)
to evaluate the effciency of selected feature subset and thus subjugating the disadvan-
tage of flter model. General framework of wrapper model in FS is shown in Figure
2.2. Embedded model merges FS and classifcation into one process (Stańczyk 2015).

FIGURE 2.2 Framework of wrapper model for feature selection.

30 Handbook of Machine Learning

Recent advancements of enormous data collections mechanism had further intri-

cated the FS technique more complex. Searching the data set for choosing optimal
feature subset exhaustively is pragmatically not feasible. Many researchers have
explored a variety of search mechanisms for FS algorithms (Lee et al. 2017) broadly
categorized as optimal, heuristic and randomized. Predominant FS techniques suf-
fer from huge computational cost and local optimum problems. However, NIFS
algorithms have received special attention among research community due to their
global search prospect. Due to their effciency and performance, NIFS algorithms
are utilized for various applications in machine learning, including classifcation,
clustering, regression, pattern recognition and image processing (Abualigah et al.
2018; Gupta et al. 2019; Gokulnath & Shantharajah 2019; Mafarja & Mirjalili 2018;
Zhang et al. 2018; Tran et al. 2017). Recent literature using NIFS for the classifcation
of EMG signals gained a prominent attention (Wu et al. 2016; Oskoei & Hu 2006;
Sharma et al. 2019). This chapter essentially concentrates on analyzing the NIFS
performance effciency of whale optimization algorithm (WOA), genetic algorithm
(GA), bat algorithm (BA), frefy optimization algorithm (FA) and particle swarm
optimization (PSO) in classifying normal and abnormal EMG signals. The consid-
ered NIFS algorithms fall under wrapper model since these algorithms use learning
model to assess the performance of selected feature subset, and further they exploit
the advantages of the wrapper models effectively.
Finally, in this chapter, ELM is constructed for evaluating the performance of
reduced subset of extracted TF features from EMG signals using NIFS techniques.
The traditional neural network uses backpropagation for learning; instead, single-
layer feed-forward neural network (SLFN) is employed for learning in ELM, neces-
sitating many user-defned parameters leading to over-ftting, and can stick to local
optimum (Huang et al. 2015). ELM also takes advantage of rapid learning and better
generalization performance.

2.2 DATA SET

The ALS, myopathy and normal signals were collected from the standard EMGLAB
database (Nikolic 2001). In this chapter, 50 normal EMG signals, 50 EMG signals
(myopathy) and 50 EMG signals (ALS) were chosen for assessing NIFS performance.
The brachial biceps muscle region was preferred for this work, and 23437.5 Hz fre-
quency sampling rate of EMG signals was acquired. Figure 2.3a–c shows the sample
EMG signals (ALS, myopathy and normal).

2.3 FEATURE EXTRACTION

In this chapter, Wigner-Ville-transformed TF are used for the extraction of effcient
features of abnormal (ALS and myopathy) and normal EMG signals. The Wigner-
Ville transformation technique distributes instantaneous power and energy spectrum
with respect to time and frequency. Further, it provides multivariate high resolu-
tion of frequency as well as time for transformed images (Qian & Chen 1994). The
Wigner-Ville transformation is expressed in equation (2.1),
Random documents with unrelated
content Scribd suggests to you:
 Agnese, and which was published in Paris in 1875 as a Portulan
de Charles Quint. (Cf. Vol. II. p. 445.)

[770] There is a fac-simile of Ziegler’s map in Vol. II. 434; also in

Goldsmid’s ed. of Hakluyt (Edinb., 1885), and in Nordenskjöld’s
Vega, i. 52.

[771] The map (1551) of Gemma Frisius in Apian is much the

same.

[772] In the Basle ed. of the Historia de Gentium. Cf. Nordenskjöld’s

Vega, vol. i., who says that the map originally appeared in
Magnus’s Auslegung und Verklarung der Neuen Mappen von den
Alten Goettenreich (Venice, 1539); and is different from the map
which appeared in the intermediate edition of 1555 at Rome, a
part of which is also annexed.

[773] The same is done in the Ptolemy of 1548 (Venice). There is a

fac-simile in Nordenskjöld’s Studien, p. 35.

[774] See Vol. IV. p. 84.

[775] We find it in the Nancy globe of about 1540 (see Vol. IV. p.
81); in the Mercator gores of 1541 (Vol. II. p. 177); and in the
Ruscelli map of 1544 (Vol. II. p. 432), where Greenland
(Grotlandia) is simply a neck connecting Europe with America;
and in Gastaldi “Carta Marina,” in the Italian Ptolemy of 1548,
where it is a protuberance on a similar neck (see Vol. II. 435; IV.
43; and Nordenskjöld’s Studien, 43). The Rotz map of 1542
seems to be based on the same material used by Mercator in his
gores, but he adds a new confusion in calling Greenland the
“Cost of Labrador.” Cf. Winsor’s Kohl Maps, no. 104. The
“Grutlandia” of the Vopellio map of 1556 is also continuous with
Labrador (see Vol. II. 436; IV. 90).

[776] See Vol. IV. pp. 42, 82.

[777] In the edition of 1562, which repeated the map, the

cartographer Moletta (Moletius) testified that its geography had
been confirmed “by letters and marine charts sent to us from
divers parts.”

[778] Winsor’s Bibliog. of Ptolemy, sub anno 1561.

[779] Lok’s map of 1582 calls it “Groetland,” the landfall of “Jac.
Scolvus,” the Pole. Cf. Vol. III. 40.

[780] For Mercator’s map, see Vol. II. 452; IV. 94, 373. Ortelius’
separate map of Scandia is much the same. It is the same with
the map of Phillipus Gallæus, dated 1574, but published at
Antwerp in 1585 in the Theatri orbis terrarum Enchiridion.
Gilbert’s map in 1576 omits the “Grocland” (Vol. III. 203). Both
features, however, are preserved in the Judæis of 1593 (Vol. IV.
97), in the Wytfliet of 1597 (Vol. II. 459), in Wolfe’s Linschoten in
1598 (Vol. III. 101), and in Quadus in 1600 (Vol. IV. 101). In the
Zaltière map of 1566 (Vol. II. 451; IV. 93), in the Porcacchi map
of 1572 (Vol. II. 96, 453; IV. 96), and in that of Johannes
Martines of 1578, the features are too indefinite for recognition.
Lelewel (i. pl. 7) gives a Spanish mappemonde of 1573.

[781] In fac-simile in Nordenskjöld’s Vega, i. 247.

[782] Vol. III p. 98.

[783] A paper by H. Rink in the Geografisk Tidskrift (viii. 139)

entitled “Ostgrönländerne i deres Forhold till Vestgrönländerne og
de övrige Eskimostammer,” is accompanied by drafts of the map
of G. Tholacius, 1606, and of Th. Thorlacius, 1668-69,—the latter
placing East Bygd on the east coast near the south end. K. J. V.
Steenstrup, on Osterbygden in Geog. Tidskrift, viii. 123, gives
facsimiles of maps of Jovis Carolus in 1634; of Hendrick Doncker
in 1669. Sketches of maps by Johannes Meyer in 1652, and by
Hendrick Doncker in 1666, are also given in the Geografisk
Tidskrift, viii. (1885), pl. 5.

[784] Voyages des Pais Septentrionaux,—a very popular book.

[785] Chips from a German Workshop, i. 327.

[786] Archæological Tour, p. 202.

[787] The earliest fixed date for the founding of Tenochtitlan

(Mexico city) is 1325. Brasseur tells us that Carlos de Sigüenza y
Gongora made the first chronological table of ancient Mexican
dates, which was used by Boturini, and was improved by Leon y
Gama,—the same which Bustamante has inserted in his edition of
Gomara. Gallatin (Amer. Ethnol. Soc. Trans., i.) gave a composite
table of events by dates before the Conquest, which is followed in
 Brantz Mayer’s Mexico as it was, i. 97. Ed. Madier de Montjau, in
his Chronologie hiéroglyphico-phonétique des Rois Astéques de
1352 à 1522, takes issue with Ramirez on some points.

[788] Bancroft (v. 199) gives references to those writers who have
discussed this question of giants. Bandelier’s references are more
in detail (Arch. Tour, p. 201). Short (p. 233) borrows largely the
list in Bancroft. The enumeration includes nearly all the old
writers. Acosta finds confirmation in bones of incredible
largeness, often found in his day, and then supposed to be
human. Modern zoölogists say they were those of the Mastodon.
Howarth, Mammoth and the Flood, 297.

[789] See Native Races, ii. 117; v. 24, 27.

[790] Sometimes it is said they came from the Antilles, or beyond,

easterly, and that an off-shoot of the same people appeared to
the early French, explorers as the Natchez Indians. We have, of
course, offered to us a choice of theories in the belief that the
Maya civilization came from the westward by the island route
from Asia. This misty history is nothing without alternatives, and
there are a plenty of writers who dogmatize about them.

[791] Constituciones diocesanas del obispado de Chiappas (Rome,

1702).

[792] Nat. Races, v. 160.

[793] Hist. Nations Civilisées, i. 37, 150, etc. Popul Vuh, introd., sec.
v. Bancroft relates the Votan myth, with references, in Nat.
Races, iii. 450. Brasseur identifies the Votanites with the Colhuas,
as the builders of Palenqué, the founders of Xibalba, and thinks a
branch of them wandered south to Peru. There are some stories
of even pre-Votan days, under Igh and Imox. Cf. H. De
Charency’s “Myth d’Imos,” in the Annales de philosophie
Chrétienne, 1872-73, and references in Bancroft, v. 164, 231.

[794] Native Races, ii. 121, etc.

[795] Bancroft (v. 236) points to Bradford, Squier, Tylor, Viollet-le-

Duc, Bartlett, and Müller, with Brasseur in a qualified way, as in
the main agreeing in this early disjointing of the Nashua stock, by
which the Maya was formed through separation from the older
race.
[796] Enforced, for instance, by one of the best of the later Mexican
writers, Orozco y Berra, in his Geografía de las lenguas y Carta
Ethnografica de México (Mexico, 1865).

[797] Tylor, Anahuac, 189, and his Early Hist. Mankind, 184. Orozco
y Berra, Geog., 124. Bancroft, v. 169, note. The word Maya was
first heard by Columbus in his fourth voyage, 1503-4. We
sometimes find it written Mayab. It is usual to class the people of
Yucatan, and even the Quiché-Cakchiquels of Guatemala and
those of Nicaragua, under the comprehensive term of Maya, as
distinct from the Nahua people farther north.

[798] Nat. Races, v. 186.

[799] Brinton, with his view of myths, speaks of the attempt of the
Abbé Brasseur to make Xibalba an ancient kingdom, with
Palenqué as its capital, as utterly unsupported and wildly
hypothetical (Myths, 251).

[800] Perhaps by Gucumatz (who is identified by some with

Quetzalcoatl), leading the Tzequiles, who are said to have
appeared from somewhere during one of Votan’s absences, and
to have grown into power among the Chanes, or Votan’s people,
till they made Tulan, where they lived, too powerful for the
Votanites. Bancroft (v. 187) holds this view against Brasseur.

[801] Perhaps Ococingo, or Copan, as Bancroft conjectures (v.

187).

[802] As Sahagún calls it, meaning, as Bancroft suggests, Tabasco.

[803] Short (p. 248) points out that the linguistic researches of
Orozco y Berra (Geografía de las Lenguas de México, 1-76) seem
to confirm this.

[804] See p. 158.

[805] Kirk says (Prescott’s Mexico): “Confusion arises from the

name of Chichimec, originally that of a single tribe, and
subsequently of its many offshoots, being also used to designate
successive hordes of whatever race.” Some have seen in the
Waiknas of the Mosquito Coast, and in the Caribs generally,
descendants of these Chichimecs who have kept to their old
social level. The Caribs, on other authority, came originally from
 the stock of the Tupis and Guaranis, who occupied the region
south of the Amazon, and in Columbus’s time they were scattered
in Darien and Honduras, along the northern regions of South
America, and in some of the Antilles (Von Martius, Beiträge sur
Ethnographie and Sprachenkunde Amerika’s zumal Brasilìens,
Leipzig, 1867). Bancroft (ii. 126) gives the etymology of
Chichimec and of other tribal designations. Cf. Buschmann’s
Ueber die Aztekischen Ortsnamen (Berlin, 1853). Bandelier
(Archæol. Tour, 200; Peabody Mus. Repts., ii. 393) says he fails
to discover in the word anything more than a general term,
signifying a savage, a hunter, or a warrior, Chichimecos, applied
to roving tribes. Brasseur says that Mexican tradition applies the
term Chichimecs generically to the first occupants of the New
World.

[806] These names wander and exchange consonants provokingly,

and it may be enough to give alphabetically a list comprised of
those in Prichard (Nat. Hist. Man) and Orozco y Berra
(Geografía), with some help from Gallatin in the American Ethno.
Soc. Trans., i., and other groupers of the ethnological traces:
Chinantecs, Chatinos, Cohuixcas, Chontales, Colhuas, Coras,
Cuitatecs, Chichimecs, Cuextecas (Guaxtecas, Huastecs),
Mazetecs, Mazahuas, Michinacas, Miztecs, Nonohualcas, Olmecs,
Otomís, Papabucos, Quinames, Soltecos, Totonacs, Triquis,
Tepanecs, Tarascos, Xicalancas, Zapotecs. It is not unlikely the
same people may be here mentioned under different names. The
diversity of opinions respecting the future of these vapory
existences is seen in Bancroft’s collation (v. 202). Torquemada
tells us about all that we know of the Totonacs, who claim to
have been the builders of Teotihuacan. Bancroft gives references
(v. 204) for the Totonacs, (p. 206) for the Otomís, (p. 207) for
the Mistecs and Zapotecs, and (p. 208) for the Huastecs.

[807] Bancroft, ii. 97. Brasseur, Nat. Civ., i. ch. 4, and his Palenqué
ch. 3.

[808] Called Huehue-Tlapallan, as Brasseur would have it.

[809] Following Motolinía and other early writers.

[810] Native Races, v. 219, 616.

[811] Bandelier, Archæol. Tour, 253.

[812] Kingsborough, ix. 206, 460; Veytia, i. 155, 163. Of the
Quetzalcoatl myth there are references elsewhere. P. J. J.
Valentini has made a study of the early Mexican ethnology and
history in his “Olmecas and Tultecas,” translated by S. Salisbury,
Jr., and printed in the Amer. Antiq. Soc. Proc., Oct. 21, 1882. On
Quetzalcoatl in Cholula, see Torquemada, translated in Bancroft,
iii. 258.

[813] This wide difference covers intervening centuries, each of

which has its advocates. Short carries their coming back to the
fourth century (p. 245), but Clavigero’s date of a.d. 544 is more
commonly followed. Veytia makes it the seventh century. Bancroft
(v. 211, 214) notes the diversity of views.

[814] Bancroft (v. 322) in a long note collates the different

statements of the routes and sojourns in this migration. Cf. Short,
p. 259.

[815] Cf. Kirk in Prescott, i. 10. It must be confessed that it is rather

in the domain of myth than of history that we must place all that
has been written about the scattering of the Toltec people at
Babel (Bancroft, v. 19), and their finally reaching Huehue-
Tlapallan, wherever that may have been. The view long prevalent
about this American starting-point of the Nahuas, Toltecs, or
whatever designation may be given to the beginners of this myth
and history, placed it in California, but some later writers think it
worth while to give it a geographical existence in the Mississippi
Valley, and to associate it in some vague way with the
moundbuilders and their works (Short, No. Amer. of Antiq., 251,
253). There is some confusion between Huehue-Tlapallan of this
story and the Tlapallan noticed in the Spanish conquest time,
which was somewhere in the Usumacinta region, and if we
accept Tollan, Tullan, or Tula as a form of the name, the
confusion is much increased (Short, pp. 217-220). Bancroft (v.
214) says there is no sufficient data to determine the position of
Huehue-Tlapallan, but he thinks “the evidence, while not
conclusive, favors the south rather than the north” (p. 216). The
truth is, about these conflicting views of a northern or southern
origin, pretty much as Kirk puts it (Prescott, i. 18): “All that can
be said with confidence is, that neither of the opposing theories
rests on a secure and sufficient basis.” The situation of Huehue-
Tlapallan and Aztlan is very likely one and the same question, as
 looking to what was the starting-point of all the Nahua
migrations, extending over a thousand years.

[816] Bancroft, v. 217.

[817] Torquemada, Boturini, Humboldt, Brasseur, Charnay, Short,

etc.

[818] Nat. Races (v. 222).

[819] In support of the California location, Buschmann, in his Ueber

die Spuren der Aztekischen Sprache im nördlichen Mexico und
höheren Amerikanischen Norden (Berlin, 1854), finds traces of
the Mexican tongue in those of the recent California Indians.
Linguistic resemblances to the Aztec, even so far north as
Nootka, have been traced, but later philologists deny the
inferences of relationship drawn from such similarity (Bancroft, iii.
p. 612). The linguistic confusion in aboriginal California is so
great that there is a wide field for tracing likenesses (Ibid. iii.
635). In the California State Mining Bureau, Bulletin no. 1
(Sacramento, 1888), Winslow Anderson gives a description of
some desiccated human remains found in a sealed cave, which
are supposed to be Aztec. There are slight resemblances to the
Aztec in the Shoshone group of languages (Bancroft, iii. 660),
and the same author arranges all that has been said to connect
the Mexican tongue with those of New Mexico and neighboring
regions (iii. 664). Buschmann, who has given particular attention
to tracing the Aztec connections at the north, finds nothing to
warrant anything more than casual admixtures with other stocks
(Die Lautveränderung Aztekischer Wörter, Berlin, 1855, and Die
Spuren der Aztekischen Sprachen, Berlin, 1859). See Short (p.
487) for a summary.

[820] Bancroft (v. 305) cites the diverse views; so does Short to
some extent (pp. 246, 258, etc.). Cf. Brinton’s Address on “Where
was Aztlan?” p. 6; Short, 486, 490; Nadaillac, 284; Wilson’s
Prehistoric Man, i. 327.
Brinton (Myths of the New World, etc., 89; Amer. Hero. Myths,
92) holds that Aztlan is a name wholly of mythical purport, which
it would be vain to seek on the terrestrial globe. This cradle
region of the Nahuas sometimes appears as the Seven Caves
(Chicomoztoc), and Duran places them “in Teoculuacan,
otherwise called Aztlan, a country toward the north and
connected with Florida.” The Seven Caves were explained by
Sahagún as a valley, by Clavigero as a city, by Schoolcraft and
others as simply seven boats in which the first comers came from
Asia; Brasseur makes them and Aztlan the same; others find
them to be the seven cities of Cibola,—so enumerates Brinton
(Myths, 227), who thinks that the seven divisions of the Nahuas
sprung from the belief in the Seven Caves, and had in reality no
existence.
Gallatin has followed out the series of migrations in the Amer.
Ethnol. Soc. Trans., i. 162. Dawson, Fossil Men (ch. 3), gives his
comprehensive views of the main directions of these early
migrations. Brasseur follows the Nahuas (Popul Vuh, introd., sect.
ix.). Winchell (Pre-Adamites) thinks the general tendency was
from north to south. Morgan finds the origin of the Mexican tribes
in New Mexico and in the San Juan Valley (Peabody Mus. Rept.,
xii. 553. Cf. his article in the North Am. Rev., Oct., 1869).
Humboldt (Views of Nature, 207) touches the Aztec wanderings.
There are two well-known Aztec migration maps, first
published in F. G. Carreri’s Giro del Mondo; in English as “Voyage
round the world,” in Churchill’s Voyages, vol. iv., concerning which
see Bancroft, ii. 543; iii. 68, 69; Short, 262, 431, 433; Prescott,
iii. 364, 382. Orozco y Berra (Hist. Antiq. de Mexico, iii. 61) says
that these maps follow one another, and are not different records
of the same progress. Humboldt (Vues, etc., ii. 176) gives an
interpretation of them in accordance with Sigüenza’s views, which
is the one usually followed, and Bancroft (v. 324) epitomizes it.
Ramirez says that the copies reproduced in Humboldt, Clavigero,
and Kingsborough are not so correct as the engraving given in
Garcia y Cubas’s Atlas geogrâfico, estadistico e histórico de la
Republica Mejicana (April, 1858). Bancroft (ii. 544) gives it as
reproduced by Ramirez. It is also in the Mexican edition of
Prescott, and in Schoolcraft’s Indian Tribes. Cf. Delafield’s Inquiry
(N. Y., 1839) and Léon de Rosny’s Les doc. écrits de l’antiq. Amér.
(Paris, 1882). The original is preserved in the Museo Nacional of
Mexico. A palm-tree on the map, near Aztlan, has pointed some
of the arguments in favor of a southern position for that place,
but Ramirez says it is but a part of a hieroglyphic name, and has
no reference to the climate of Aztlan (Short, p. 266). F. Von
Hellwald printed a paper on “American migrations,” with notes by
Professor Henry, in the Smithsonian Report, 1866, pp. 328-345.
Short defines as “altogether the most enlightened treatment of
the subject” the paper of John H. Becker, “Migrations des
 Nahuas,” in the Compte rendu, Congrès des Américanistes
(Luxembourg, 1877), i. 325. This paper finds an identification of
the Tulan Zuiva of the Quichés, the Huehue-Tlapallan of the
Toltecs, the Amaquemecan of the Chichimecs, and the Oztotlan
(Aztlan) of the Aztecs in The valleys of the Rio Grande del Norte
and Rio Colorado, as was Morgan’s view. Short (p. 249)
summarizes his paper. Bancroft (v. 289) shows the diversity of
views respecting Amaquemecan.

[821] Native Races, v. 167, recapitulates the proofs against the

northern theory. J. R. Bartlett, Personal Narrative, ii. 283, finds no
evidence for it. The successive sites of their sojourns as they
passed on their journeys are given as Tlapallan, Tlacutzin,
Tlapallanco, Jalisco, Atenco, Iztachnexuca, Tollatzinco, Tollan or
Tula,—the last, says Bancroft, apparently in Chiapas. If there was
not such confusion respecting the old geography, these names
might decide the question.

[822] Writers usually place the beginnings of credible history at

about this period. Brasseur and the class of writers who are easily
lifted on their imagination talk about traces of a settled
government being discernible at periods which they place a
thousand years before Christ.

[823] References in Bancroft, v. 247, with Brasseur for the main

dependence, in his use of the Codex Chimalpòpoca and the
Memorial de Colhuacan.

[824] Charnay (Eng. trans., ch. 8 and 9) calls it a rival city of Tula or
Tollan, rebuilt by the Chichimecs on the ruins of a Toltec city.

[825] If one wants the details of all this, he can read it in Veytia,
Brasseur (Nat. Civilisées and Palenqué, ch. viii.), and Bancroft,
the latter giving references (v. 285).

[826] It is frequently stated that there was a segregated migration

to Central America. Bancroft (v. 168, 285), who collates the
authorities, finds nothing of the kind implied. He thinks the mass
remained in Anáhuac. The old view as expressed by Prescott (i.
14) was that “much the greater number probably spread over the
region of Central America and the neighboring isles, and the
traveller now speculates on the majestic ruins of Mitla and
Palenqué as possibly the work of this extraordinary people.” Kirk,
as Prescott’s editor, refers to the labors of Orozco y Berra
 (Geografía de las Lenguas de México, 122), followed by Tylor,
(Anahuac, 189) as establishing the more recent view that this
southern architecture, “though of a far higher grade, was long
anterior to the Toltec dominion.”

[827] Amer. Ethno. Soc. Trans., i.

[828] Bancroft (v. 287) says: “It is probable that the name Toltec, a
title of distinction rather than a national name, was never applied
at all to the common people.”

[829] Brinton’s main statement is in his Were the Toltecs an historic

nationality? Read before the American Philosophical Society, Sept.
2, 1887 (Phila., 1887); published also in their Proceedings, 1887,
p. 229. Cf. also Brinton’s Amer. Hero. Myths (Phil., 1882), p. 86,
where he throws discredit on the existence of the alleged Toltec
king Quetzalcoatl (whom Sahagún keeps distinct from the
mythical demi-god); and earlier, in his Myths of the New World
(p. 29), he had suggested that the name Toltec might have “a
merely mythical signification.” Charnay, who makes the Toltecs a
Nahuan tribe, had defended their historical status in a paper on
“La Civilisation Tolteque,” in the Revue d’Ethnographie (iv., 1885);
and again, two years later, in the same periodical, he reviewed
adversely Brinton’s arguments. (Cf. Saturday Review, lxiii. 843.)
Otto Stoll, in his Guatemala, Reisen und Schilderungen (Leipzig,
1886), is another who rejects the old theory.

[830] Archæol. Tour, 253.

[831] Archæol. Tour, 7. Sahagún identifies the Toltecs with the

“giants,” and if these were the degraded descendants of the
followers of Votan, Sahagún thus earlier established the same
identity.

[832] Archæol. Tour, 191. The fact that the names which we
associate with the Toltecs are Nahua, only means that Nahua
writers have transmitted them, as Bandelier thinks. Cf. also
Bandelier’s citation in the Peabody Mus. Reports, vol. ii. 388,
where he speaks of our information regarding the Toltecs as
“limited and obscure.” He thinks it beyond question that they
were Nahuas; and the fact that their division of time corresponds
with the system found in Yucatan, Guatemala, etc., with other
evidences of myths and legends, leads him to believe that the
aborigines of more southern regions were, if not descendants, at
 least of the same stock with the Toltecs, and that we are justified
in studying them to learn what the Toltecs were. He finds that
Veytia, in his account of the Toltecs, beside depending on
Sahagún and Torquemada, finds a chief source in Ixtlilxochitl, and
locates Huehue-Tlapallan in the north; and Veytia’s statements
reappear in Clavigero.
The best narratives of the Toltec history are those in Veytia,
Historia Antigua de Méjico (Mexico, 1806); Brasseur’s Hist.
Nations Civilisées (vol. i.), and his introduction to his Popul Vuh;
and Bancroft (v. ch. 3 and 4): but we must look to Ixtlilxochitl,
Torquemada, Sahagún, and the others, if we wish to study the
sources. In such a study we shall encounter vexatious problems
enough. It is practically impossible to arrange chronologically
what Ixtlilxochitl says that he got from the picture-writings which
he interpreted. Bancroft (v. 209) does the best he can to give it a
forced perspicuity. Wilson (Prehisoric Man, i. 245) not inaptly
says: “The history of the Toltecs and their ruined edifices stands
on the border line of romance and fable, like that of the ruined
builders of Carnac and Avebury.”

[833] Short (page 255) points out that Bancroft unadvisedly looks
upon these Chichimecs as of Nahua stock, according to the
common belief. Short thinks that Pimentel (Lenguas indigenas de
México, published in 1862) has conclusively shown that the
Chichimecs did not originally speak the Nahua tongue, but
subsequently adopted it. Short (page 256) thinks, after collating
the evidence, that it is impossible to determine whence or how
they came to Anáhuac.

[834] Bancroft, v. 292, gives the different views. Cf. Kirk in Prescott,
i. 16.

[835] These events are usually one thing or another, according to

the original source which you accept, as Bancroft shows (v. 303).
The story of the text is as good as any, and is in the main borne
out by the other narratives.

[836] Bancroft, v. 308. Cf., on the arrival of the Mexicans in the

valley, Bandelier (Peabody Mus. Reports, ii. 398) and his
references.

[837] Prescott, i., introduction ch. 6, tells the story of their golden
age.
[838] Cf. the map in Lucien Biart’s Les Aztèques (Paris, 1885).
Prescott says the maps in Clavigero, Lopez, and Robertson defy
“equally topography and history.” Cf. note on plans of the city and
valley in Vol. II. pp. 364, 369, 374, to which may be added, as
showing diversified views, those in Stevens’s Herrera (London,
1740), vol. ii.; Bordone’s Libro (1528); Icazbalceta’s Coll. de
docs., i. 390; and the Eng. translation of Cortes’ despatches, 333.

[839] This is placed a.d. 1325. Cf. references in Bancroft (v. 346).

[840] On the conquest of the Tecpanecas by the Mexicans, see the

references in Bandelier (Peabody Mus. Reports, ii. 412).

[841] For details of the period of the Chichimec ascendency, see

Bancroft (v. ch. 5-7), Brasseur (Nat. Civil. ii.), and the authorities
plentifully cited in Bancroft.

[842] On the nature of the Mexican confederacy see Bandelier

(Peabody Mus. Reports, ii. 416). He enumerates the authorities
upon the point that no one of the allied tribes exercised any
powers over the others beyond the exclusive military direction of
the Mexicans proper (Peabody Mus. Reports, ii. 559). Orozco y
Berra (Geografía, etc.) claims that there was a tendency to
assimilate the conquered people to the Mexican conditions.
Bandelier claims that “no attempt, either direct or implied, was
made to assimilate or incorporate them.” He urges that nowhere
on the march to Mexico did Cortés fall in with Mexican rulers of
subjected tribes. It does not seem to be clear in all cases whether
it was before or after the confederation was formed, or whether it
was by the Mexicans or Tezcucans that Tecpaneca, Xochimilca,
Cuitlahuac, Chalco, Acolhuacan, and Quauhnahuac, were
conquered. Cf. Bandelier in Peabody Mus. Reports, ii. 691. As to
the tributaries, see Ibid. 695.

[843] Cf. Brasseur’s Nations Civ. ii. 457, on Tezcuco in its palmy
days.

[844] Sometimes written Mochtheuzema, Moktezema. The Aztec

Montezuma must not, as is contended, be confounded with the
hero-god of the New Mexicans. Cf. Bancroft, iii. 77, 171; Brinton’s
Myths, 190; Schoolcraft’s Ind. Tribes, iv. 73; Tylor’s Prim. Culture,
ii. 384; Short, 333.
[845] This has induced some historians to call these wars “holy
wars.” Bandelier discredits wholly the common view, that wars
were undertaken to secure victims for the sacrificial stone
(Archæol. Tour, 24). But in another place (Peabody Mus. Reports,
ii. 128) he says: “War was required for the purpose of obtaining
human victims, their religion demanding human sacrifices at least
eighteen times every year.”

[846] As to these carvings, which have not yet wholly disappeared,

see Peabody Mus. Reports, ii. 677, 678. There is a series of
alleged portraits of the Mexican kings in Carbajal-Espinosa’s Hist.
de Mexico (Mexico, 1862). See pictures of Montezuma II. in Vol.
II. 361, 363, and that in Ranking, p. 313.

[847] Bancroft (v. 466) enumerates the great variety of such proofs
of disaster, and gives references (p. 469). Cf. Prescott, i. p. 309.

[848] Tezozomoc (cap. 106) gives the description of the first

bringing of the news to Montezuma of the arrival of the Spaniards
on the coast.

[849] Brinton’s Amer. Hero Myths, 139, etc. See, on the prevalence
of the idea of the return at some time of the hero-god, Brinton’s
Myths of the New World, p. 160. “We must remember,” he says,
“that a fiction built on an idea is infinitely more tenacious of life
than a story founded on fact.” Brinton (Myths, 188) gathers from
Gomara, Cogolludo, Villagutierre, and others, instances to show
how prevalent in America was the presentiment of the arrival and
domination of a white race,—a belief still prevailing among their
descendants of the middle regions of America who watch for the
coming of Montezuma (Ibid. p. 190). Brinton does not seem to
recognize the view held by many that the Montezuma of the
Aztecs was quite a different being from the demi-god of the
Pueblas of New Mexico.

[850] It is not easy to reconcile the conflicting statements of the

native historians respecting the course of events during the Aztec
supremacy, such is the mutual jealousy of the Mexican and
Tezcucan writers. Brasseur has satisfied himself of the
authenticity of a certain sequence and character of events
(Nations Civilisées), and Bancroft simply follows him (v. 401).
Veytia is occupied more with the Tezcucans than with the Aztecs.
The condensed sketch here given follows the main lines of the
collated records. We find good pictures of the later history of
 Mexico and Tlascala, before the Spaniards came, in Prescott (i.
book 2d, ch. vi., and book 3d, ch. ii.). Bancroft (v. ch. 10) with his
narrative and references helps us out with the somewhat
monotonous details of all the districts of Mexico which were
outside the dominance of the Mexican valley, as of Cholula,
Tlascala, Michoacan, and Oajaca, with the Miztecs and Zapotecs,
inhabiting this last province.

[851] Bancroft (v. 543-553).

[852] It is so held by Stephens, Waldeck, Mayer, Prichard, Ternaux-

Compans, not to name others.

[853] Vol. v. 617.

[854] The Maya calendar and astronomical system, as the basis of

the Maya chronology, is explained in the version which Perez gave
into Spanish of a Maya manuscript (translated into English by
Stephens in his Yucatan), and which Valentini has used in his
“Katunes of Maya History,” in the Amer. Antiq. Soc. Proc., Oct.
1879. On the difficulties of the subject see Brasseur’s Nations
Civilisées (ii. ch. 1). Cf. also his Landa, section xxxix., and page
366, from the “Cronologia antigua de Yucatan.” Cf. further, Cyrus
Thomas’s MS. Troano, ch. 2, and Powell’s Third Report Bur. of
Ethn., pp. xxx and 3; Ancona’s Yucatan, ch. xi.; Bancroft’s Nat.
Races, ii. ch. 24, with references; Short, ch. 9; Brinton’s Maya
Chronicles, introduction, p. 50.

[855] Bancroft (v. 624) epitomizes the Perez manuscript given by

Stephens, the sole source of this Totul Xiu legendary.

[856] Brasseur’s Nations Civilisées (i., ii.), with the Perez

manuscript, and Landa’s Relacion, are the sufficient source of the
Yucatan history. Bancroft’s last chapter of his fifth volume
summarizes it.

[857] See Vol. II. p. 402.

[858] See Vol. II. p. 397.

[859] Central America, ii. 452.

[860] See Vol. II. p. 414.

[861] See Vol. II. p. 343.

[862] See Vol. II. p. 412.

[863] See Vol. II. p. 417. Cf. Prescott’s Mexico, i. 50; Bancroft (Nat.
Races, ii. ch. 14) epitomizes the information on the laws and
courts of the Nahua; Bandelier (Peabody Mus. Repts., ii. 446),
referring to Zurita’s Report, which he characterizes as marked for
perspicacity, deep knowledge, and honest judgment, speaks of it
as embodying the experience of nearly twenty years,—eleven of
which were passed in Mexico,—and in which the author gave
answers to inquiries put by the king. “If we could obtain,” says
Bandelier, “all the answers given to these questions from all parts
of Spanish America, and all as elaborate and truthful as those of
Zurita, Palacio, and Ondegardo, our knowledge of the aboriginal
history and ethnology of Spanish America would be much
advanced.” Zurita’s Report in a French translation is in Ternaux-
Compans’ Collection; the original is in Pacheco’s Docs. inéditos,
but in a mutilated text.

[864] See Vol. II. p. 346.

[865] It is much we owe to the twelve Franciscan friars who on May

13, 1524, landed in Mexico to convert and defend the natives. It
is from their writings that we must draw a large part of our
knowledge respecting the Indian character, condition, and history.
These Christian apostles were Martin de Valencia, Francisco de
Soto, Martin de Coruña, Juan Xuares, Antonio de Ciudad Rodrigo,
Toribio de Benavente, Garcia de Cisneros, Luis de Fuensalida,
Juan de Ribas, Francisco Ximenez, Andrés de Cordoba, Juan de
Palos.
From the Historia of Las Casas, particularly from that part of it
called Apologética historia, we can also derive some help. (Cf.
Vol. II. p. 340.)

[866] Brasseur, Bib. Mex.-Guat., p. 147; Leclerc, p. 168.

[867] Herrera is furthermore the source of much that we read in

later works concerning the native religion and habits of life. See
Vol. II. p. 67.

[868] Cf. Vol. II. p. 418.

[869] Anales del Museo Nacional, iii. 4, 120; Brinton’s Am. Hero
Myths, 78. Bandelier, in N. Y. Hist. Soc. Proc., November, 1879,
used a portion of the MS. as printed by Sir Thomas Phillipps
 (Amer. Antiq. Soc. Proc., i. 115) under the title of Historia de los
Yndios Mexicanos, por Juan de Tovar; Cura et impensis Dni
Thomæ Phillipps, Bart. (privately printed at Middle Hill, 1860. See
Squier Catalogue, no. 1417). The document is translated by
Henry Phillipps, Jr., in the Proc. Amer. Philosophical Soc. (Philad.),
xxi. 616.

[870] Vol. II. p. 419. Brasseur de Bourbourg’s Bibl. Mex.-Guat., p.

59. He used a MS. copy in the Force collection.

[871] This is true of Acosta and Davila Padilla. The bibliography of

Acosta has been given elsewhere (Vol. II. p. 420). His books v.,
vi., and vii. cover the ancient history of the country. He used the
MSS. of Duran (Brasseur, Bibl. Mex.-Guat., p. 2), and his
correspondence with Tobar, preserved in the Lenox library, has
been edited by Icazbalceta in his Don Fray Zumárraga (Mexico,
1881). Of the Provincia de Santiago and the Varia historia of
Davila Padilla, the bibliography has been told in another place.
(Cf. Vol. II. pp. 399-400[; Sabin, v. 18780-1; Brasseur de
Bourbourg’s Bibl. Mex.-Guat., p. 53; Del Monte Library, no. 126.)
Ternaux was not wrong in ascribing great value to the books.]

[872] Peter of Ghent. Cf. Vol. II. p. 417.

[873] Chronica Compendiosissima ab exordio mundi per Amandum

Zierixcensem, adjectæ sunt epistolæ ex nova maris Oceani
Hispania ad nos transmissæ (Antwerp, 1534). The subjoined
letters here mentioned are, beside that referred to, two others
written in Mexico (1531), by Martin of Valencia and Bishop
Zumárraga (Sabin, i. no. 994; Quaritch, 362, no. 28583, £7 10).
Icazbalceta (Bib. Mex. del Siglo xvi., i. p. 33) gives a long account
of Gante. There is a French version of the letter in Ternaux’s
Collection.

[874] See Vol. II. p. 397. Cf. Prescott, ii. 95. The first part of the
Historia is on the religious rites of the natives; the second on
their conversion to Christianity; the third on their chronology, etc.

[875] Cf. Icazbalceta’s Bibl. Mexicana, p. 220, with references;

Pilling’s Proof-sheets, no. 2600, etc.

[876] Pilling, no. 2817, etc.

[877] Properly, Bernardino Ribeira; named from his birthplace,
Sahagún, in Spain. Chavero’s Sahagún (Mexico, 1877).

[878] A few data can be added to the account of Sahagún given in

Vol. II. p. 415. J. F. Ramirez completes the bibliography of
Sahagún in the Boletin de la Real Academia de la Historia de
Madrid, vi. 85 (1885). Icazbalceta, having told the story of
Sahagún’s life in his edition of Mendieta’s Hist. Eclesiastica
Indiana (México, 1870), has given an extended critical and
bibliographical account in his Bibliografía Mexicana (México,
1886), vol. i. 247-308. Other bibliographical detail can be gleaned
from Pilling’s Proof-sheets, p. 677, etc.; Icazbalceta’s Apuntes;
Beristain’s Biblioteca; the Bibliotheca Mexicana of Ramirez. The
list in Adolfo Llanos’s Sahagún y su historia de México (Museo
Nac. de Méx. Anales, iii., pt. 3, p. 71) is based chiefly on Alfredo
Chavero’s Sahagún (México, 1877). Brasseur de Bourbourg, in his
Palenqué (ch. 5), has explained the importance of what Brevoort
calls Sahagún’s “great encyclopædia of the Mexican Empire.”
Rosny (Les documents écrits de l’Antiquité Américaine, p. 69)
speaks of seeing a copy of the Historia in Madrid, accompanied
by remarkable Aztec pictures. Bancroft, referring to the defective
texts of Sahagún in Kingsborough and Bustamante, says:
“Fortunately what is missing in one I have always found in the
other.” He further speaks of the work of Sahagún as “the most
complete and comprehensive, so far as aboriginal history is
concerned, furnishing an immense mass of material, drawn from
native sources, very badly arranged and written.” Eleven books of
Sahagún are given to the social institutions of the natives, and
but one to the conquest. Jourdanet’s edition is mentioned
elsewhere (Vol. II.).

[879] See Vol. II. p. 421.

[880] Those who used him most, like Clavigero and Brasseur de
Bourbourg, complain of this. Torquemada, says Bandelier
(Peabody Mus. Repts. ii. 119), “notwithstanding his
unquestionable credulity, is extremely important on all questions
of Mexican antiquities.”

[881] Amer. Antiq. Soc. Proc., n. s., i. 105.

[882] Cf. Vol. II. 417; Prescott, i. 13, 163, 193, 196; Bancroft, Nat.
Races, v. 147; Wilson’s Prehistoric Man, i. 325. It must be
confessed that with no more authority than the old Mexican
 paintings, interpreted through the understanding of old men and
their traditions, Ixtlilxochitl has not the firmest ground to walk on.
Aubin thinks that Ixtlilxochitl’s confusion and contradictions arise
from his want of patience in studying his documents; and some
part of it may doubtless have arisen from his habit, as Brasseur
says (Annales de Philosophie Chrétienne, May, 1855, p. 329), of
altering his authorities to magnify the glories of his genealogic
line. Max Müller (Chips from a German Workshop, i. 322) says of
his works: “Though we must not expect to find in them what we
are accustomed to call history, they are nevertheless of great
historical interest, as supplying the vague outlines of a distant
past, filled with migrations, wars, dynasties and revolutions, such
as were cherished in the memory of the Greeks in the time of
Solon.” In addition to his Historia Chichimeca and his Relaciones,
(both of which are given by Kingsborough, while Ternaux has
translated portions,)—the MS. of the Relaciones being in the
Mexican archives,—Ixtlilxochitl left a large mass of his manuscript
studies of the antiquities, often repetitionary in substance. Some
are found in the compilation made in Mexico by Figueroa in 1792,
by order of the Spanish government (Prescott, i. 193). Some
were in the Ramirez collection. Quaritch (MS. Collections, Jan.,
1888, no. 136) held one from that collection, dated about 1680,
at £16, called Sumaria Relacion, which concerned the ancient
Chichimecs. Those which are best known are a Historia de la
Nueva España, or Historia del Reyno de Tezcuco, and a Historia
de Nuestra Señora de Guadalupe, if this last is by him.

[883] Annales de Philosophie Chrétienne, May, 1855, p. 326.

[884] In his Quatre Lettres, p. 24, he calls it the sacred book of the
Toltecs. “C’est le Livre divin lui-même, c’est le Teoamoxtli.”

[885] Brasseur’s Lettres à M. le due de Valmy, Lettre seconde.

[886] Catálogo, pp. 17, 18.

[887] Brasseur, Bibl. Mex. Guat., p. 47; Pinart-Brasseur Catal., no.

237.

[888] It has been announced that Bandelier is engaged in a new

translation of The Annals of Quauhtitlan for Brinton’s Aboriginal
Literature series. Cf. Bancroft, iii. 57, 63, and in vol. v., where he
endeavors to patch together Brasseur’s fragments of it. Short, p.
241.
[889] Humboldt says that Sigüenza inherited Ixtlilxochitl’s
collection; and that it was preserved in the College of San Pedro
till 1759.

[890] Giro del mondo, 1699, vol. vi. Cf. Kingsborough, vol. iv.
Robertson attacked Carreri’s character for honesty, and claimed it
was a received opinion that he had never been out of Italy.
Clavigero defended Carreri. Humboldt thinks Carreri’s local
coloring shows he must have been in Mexico.

[891] Cf. the bibliog., in Vol. II., p. 425, of his Storia Antica del
Messico.

[892] We owe to him descriptions at this time of the collections of

Mendoza, of that in the Vatican, and of that at Vienna. Robertson
made an enumeration of such manuscripts; but his knowledge
was defective, and he did not know even of those at Oxford.

[893] Robertson was inclined to disparage Clavigero’s work,

asserting that he could find little in him beyond what he took
from Acosta and Herrera “except the improbable narratives and
fanciful conjectures of Torquemada and Boturini.” Clavigero
criticised Robertson, and the English historian in his later editions
replied. Prescott points out (i. 70) that Clavigero only knew
Sahagún through the medium of Torquemada and later writers.
Bancroft (Nat. Races, v. 149; Mexico, i. 700) thinks that Clavigero
“owes his reputation much more to his systematic arrangement
and clear narration of traditions that had before been greatly
confused, and to the omission of the most perplexing and
contradictory points, than to deep research or new discoveries.”

[894] See Vol. II. p. 418. Brasseur de Bourbourg’s Hist. des Nations
Civilisées, p. xxxii. Clavigero had described it.

[895] He had collected nearly 500 Mexican paintings in all. Aubin

(Notices, etc., p. 21) says that Boturini nearly exhausted the field
in his searches, and with the collection of Sigüenza he secured all
those cited by Ixtlilxochitl and the most of those concealed by the
Indians,—of which mention is made by Torquemada, Sahagún,
Valadés, Zurita, and others; and that the researches of
Bustamante, Cubas, Gondra, and others, up to 1851, had not
been able to add much of importance to what Boturini possessed.
[896] This portion of his collection has not been traced. The fact is
indeed denied.

[897] Idea de una nueva historia general de la America

septentrional (Madrid, 1746); Carter-Brown, iii. 817; Brasseur’s
Bibl. Mex.-Guat., p. 26; Field, Ind. Bibliog., no. 159; Pinart,
Catalogue, no. 134; Prescott, i. 160.

[898] Brasseur, Bibl. Mex.-Guat., p. 152.

[899] Prescott, i. 24. Harrisse, Bib. Am. Vet., calls Veytia’s the best
history of the ancient period yet (1866) written.

[900] A second ed. (Mexico, 1832) was augmented with notes and
a life of the author, by Carlos Maria de Bustamante; Field, Ind.
Bibliog., no. 909; Brasseur’s Bibl. Mex.-Guat., p. 68.

[901] Prescott, i. 133. Gama and others collected another class of

hieroglyphics, of less importance, but still interesting as
illustrating legal and administrative processes used in later times,
in the relations of the Spaniards with the natives; and still others
embracing Christian prayers, catechisms, etc., employed by the
missionaries in the religious instruction (Aubin, Notice, etc., 21).
Humboldt (vol. xiii., pl. p. 141) gives “a lawsuit in hieroglyphics.”
There was published (100 copies) at Madrid, in 1878, Pintura
del Gobernador, Alcaldes y Regidores de México, Codice en
geroglíficos Méxicanos y en lengua Castellana y Azteca, Existente
en la Biblioteca del Excmo Señor Duque de Osuna,—a legal
record of the later Spanish courts affecting the natives.

[902] Humboldt describes these collections which he knew at the

beginning of the century, speaking of José Antonio Pichardo’s as
the finest.

[903] Notice sur une collection d’antiquités Mexicaines, being an

extract from a Mémoire sur la peinture didactique et l’Écriture
figurative des Anciens Mexicains (Paris, 1851; again, 1859-1861).
Cf. papers in Revue Américaine et Orientale, 1st ser., iii., iv., and
v. Aubin says that Humboldt found that part of the Boturini
collection which had been given over to the Mexican archivists
diminished by seven eighths. He also shows how Ternaux-
Compans (Crauatés Horribles, p. 275-289), Rafael Isidro Gondra
(in Veytia, Hist. Ant. de Mex., 1836, i. 49), and Bustamante have
related the long contentions over the disposition of these relics,
and how the Academy of History at Madrid had even secured the
suppression of a similar academy among the antiquaries in
Mexico, which had been formed to develop the study of their
antiquities. It was as a sort of peace-offering that the Spanish
king now caused Veytia to be empowered to proceed with the
work which Boturini had begun. This allayed the irritation for a
while, but on Veytia’s death (1769) it broke out again, when
Gama was given possession of the collection, which he further
increased. It was at Gama’s death sold at auction, when
Humboldt bought the specimens which are now in Berlin, and
Waldeck secured others which he took to Europe. It was from
Waldeck that Aubin acquired the Boturini part of his collection.
The rest of the collection remained in Mexico, and in the main
makes a part at present of the Museo Nacional. But Aubin is a
doubtful witness.
Aubin says that he now proposed to refashion the Boturini
collection by copies where he could not procure the originals; to
add others, embracing whatever he could still find in the hands of
the native population, and what had been collected by Veytia,
Gama, and Pichardo. In 1851, when he wrote, Aubin had given
twenty years to this task, and with what results the list of his
MSS., which he appends to the account we have quoted, will
show.
These include in the native tongue:—
a. History of Mexico from a.d. 1064 to 1521, in fragments,
from Tezozomoc and from Alonso Franco, annotated by Domingo
Chimalpain (a copy).
b. Annals of Mexico, written apparently in 1528 by one who
had taken part in the defence of Mexico (an original).
c. Several historical narratives on European paper, by Domingo
Chimalpain, coming down to a.d. 1591, which have in great part
been translated by Aubin, who considers them the most
important documents which we possess.
d. A history of Colhuacan and Mexico, lacking the first leaf.
This is described as being in the handwriting of Ixtlilxochitl, and
Aubin gives the dates of its composition as 1563 and 1570. It is
what has later been known as the Codex Chimalpopoca.
e. Zapata’s history of Tlaxcalla.
f. A copy by Loaysa of an original, from which Torquemada has
copied several chapters.
[904] The chief of the Boturini acquisition he enumerates as
follows:—
a. Toltec annals on fifty leaves of European paper, cited by
Gama in his Descripcion histórica. Cf. Brasseur, Nations Civilisées,
p. lxxvi.
b. Chichimec annals, on Indian paper, six leaves, of which ten
pages consist of pictures, the original so-called Codex
Chimalpopoca, of which Gama made a copy, also in the Aubin
collection, as well as Ixtlilxochitl’s explanation of it. Aubin says
that he has used this account of Ixtlilxochitl to rectify that
historian’s blunders.
c. Codex on Indian paper, having a picture of the Emperor
Xolotl.
d. A painting on prepared skin, giving the genealogy of the
Chichimecan chiefs, accompanied by the copies made by
Pichardo and Boturini. Cf. Archives de la Soc. Amér. de France, 2d
ser., i. 283.
e. A synchronical history of Tepechpan and of Mexico, on
Indian paper, accompanied by a copy made by Pichardo and an
outline sketch of that in the Museo Nacional.
Without specifying others which Aubin enumerates, he gives as
other acquisitions the following in particular:—
a. Pichardo’s copy of a Codex Mexicanus, giving the history of
the Mexicans from their leaving Aztlan to 1590.
b. An original Mexican history from the departure from Aztlan
to 1569.
c. Fragments which had belonged to Sigüenza.

[905] Notice sur une Collection, etc., p. 12.

[906] Hist. des Nations Civilisées (i. pp. xxxi, lxxvi, etc.; cf. Müller’s
Chips, i. 317, 320, 323). Brasseur in the same place describes his
own collection; and it may be further followed in his Bibl. Mex.-
Guat., and in the Pinart Catalogue. Dr. Brinton says that we owe
much for the preservation during late years of Maya MSS. to Don
Juan Pio Perez, and that the best existing collection of them is
that of Canon Crescencio Carrillo y Ancona. José F. Ramirez (see
Vol. II. p. 398) is another recent Mexican collector, and his MSS.
have been in one place and another in the market of late years.
Quaritch’s recent catalogues reveal a number of them, including
his own MS. Catálogo de Colecciones (Jan., 1888, no. 171), and
some of his unpublished notes on Prescott, not included in those
 “notas y ecclarecimientos” appended to Navarro’s translation of
the Conquest of Mexico (Catal., 1885, no. 28,502). The several
publications of Léon de Rosny point us to scattered specimens. In
his Doc. écrits de l’Antiquité Amér. he gives the fac-simile of a
colored Aztec map. A MS. in the collection of the Corps Legislatif,
in Paris, and that of the Codex Indiæ Meridionalis are figured in
his Essai sur le déchiffrement, etc. (pl. ix, x). In the Archives de
la Soc. Amér. de France, n. s., vol. i., etc., we find plates of the
Mappe Tlotzin, and a paper of Madier de Montjau, “sur quelques
manuscrits figuratifs de l’Ancien Méxique.” Cf. also Anales del
Museo, viii.
Cf. for further mention of collections the Revue Orientale et
Américaine; Cyrus Thomas in the Am. Antiquarian, May, 1884
(vol. vi.); and the more comprehensive enumeration in the
introduction to Domenech’s Manuscrit pictographique. Orozco y
Berra, in the introduction to his Geografia de las Lenguas y Carta
Etnográfica (Mexico, 1864), speaks of the assistance he obtained
from the collections of Ramirez and of Icazbalceta.

[907] See Vol. II. p. 418.

[908] See Vol. II. p. 418. Bandelier calls this French version “utterly
unreliable.”

[909] This is Beristain’s title. Torquemada, Vetancurt, and Sigüenza

cite it as Memorias históricas; Brasseur, Bib. Mexico-Guat., p. 122.

[910] Cf. “Les Annales Méxicaines,” by Rémi Siméon in the Archives

de la Soc. Amér. de France, n. s., vol. ii.

[911] It is cited by Chavero as Codex Zumárraga.

[912] Hist. Nat. Civ., ii. 577.

[913] Aboriginal Amer. Authors, p. 29. Cf. Bandelier’s Bibliography

of Yucatan in Am. Antiq. Soc. Proc., n. s., vol. i. p. 82. Cf. the
references in Brasseur, Hist. Nat. Civ., and in Bancroft, Nat.
Races, v.

[914] Cf. Mem. of Berendt, by Brinton (Worcester, 1884).

[915] Cf. Brinton on the MSS. in the languages of Cent. America, in

Amer. Jour. of Science, xcvii. 222; and his Books of Chilan Balam,
the prophetic and historical records of the Mayas of Yucatan
 (Philad., 1882), reprinted from the Penn Monthly, March, 1882.
Cf. also the Transactions of the Philad. Numismatic and
Antiquarian Soc.

[916] This is in the alphabet adopted by the early missionaries. The

volume contains the “Books of Chilan Balam,” written “not later
than 1595,” and also the “Chac Xulub Chen,” written by a Maya
chief, Nakuk Pech, in 1562, to recount the story of the Spanish
conquest of Yucatan.

[917] This was in 1843, when Stephens made his English

translation from Pio Perez’s Spanish version, Antigua Chronologia
Yucateca; and from Stephens’s text, Brasseur gave it a French
rendering in his edition of Landa. (Cf. also his Nat. Civilisées, ii. p.
2.) Perez, who in Stephens’s opinion (Yucatan, ii. 117) was the
best Maya scholar in that country, made notes, which Valentini
published in his “Katunes of Maya History,” in the Pro. of the
Amer. Antiq. Soc., Oct., 1879 (Worcester, 1880), but they had
earlier been printed in Carrillo’s Hist. y Geog. de Yucatan (Merida,
1881). Bancroft (Nat. Races, v. 624) reprints Stephens’s text with
notes from Brasseur.
The books of Chilan Balam were used both by Cogolludo and
Lizana; and Brasseur printed some of them in the Mission
Scientifique au Méxique. They are described in Carrillo’s
Disertacion sobre la historia de lengua Maya ó Yucateca (Merida,
1870).

[918] Brasseur, Bib. Mex.-Guat., p. 30. See Vol. II. p. 429. The
Spanish title is Relacion de las Cosas de Yucatan.

[919] From the Proc. of the Amer. Philos. Soc., xxiv.

[920] Cf. Bandelier in Am. Antiq. Soc. Proc., n. s., vol. i. p. 88.

[921] The second edition was called Los tres Siglos de la

Dominacion Española en Yucatan (Campeche and Merida, 2 vols.,
1842, 1845). It was edited unsatisfactorily by Justo Sierra. Cf.
Vol. II. p. 429; Brasseur, Bib. Mex.-Guat., p. 47.
This, like Juan de Villagutierre Soto-Mayor’s Historia de la
Conquista de la Provincia de el Itza, reduccion, y progressos de la
de el Lacandon, y otras naciones de Indios Barbaros, de la
mediacion de el Reyno de Gautimala, a las Provincias de Yucatan,
en la America Septentrional (Madrid, 1701), (which, says
 Bandelier, is of importance for that part of Yucatan which has
remained unexplored), has mostly to do with the Indians under
the Spanish rule, but the books are not devoid of usefulness in
the study of the early tribes.
Of the modern comments on the Yucatan ancient history,
those of Brasseur in his Nations Civilisées are more to be trusted
than his introduction to his edition of Landa, which needs to be
taken with due recognition of his later vagaries; and Brinton has
studied their history at some length in the introduction to his
Maya Chronicles. The first volume of Eligio Ancona’s Hist. de
Yucatan covers the early period. See Vol. II. p. 429. Brinton calls
it “disappointingly superficial.” There is much that is popularly
retrospective in the various and not always stable contributions of
Dr. Le Plongeon and his wife. The last of Mrs. Le Plongeon’s
papers is one on “The Mayas, their customs, laws, religion,” in the
Mag. Amer. Hist., Aug., 1887. Bancroft’s second volume groups
the necessary references to every phase of Maya history. Cf.
Charnay, English translation, ch. 15; and Geronimo Castillo’s
Diccionario Histórico, biográfico y monumental de Yucatan
(Mérida, 1866). Of Crescencio Carrillo and his Historia Antigua de
Yucatan (Mérida, 1881), Brinton says: “I know of no other
Yucatecan who has equal enthusiasm or so just an estimate of
the antiquarian riches of his native land” (Amer. Hero Myths,
147). Bastian summarizes the history of Yucatan and Guatemala
in the second volume of his Culturländer des alten Amerika.

[922] Yucatan, ii. 79.

[923] See C. H. Berendt on the hist. docs. of Guatemala in

Smithsonian Report, 1876. There is a partial bibliography of
Guatemala in W. T. Brigham’s Guatemala the land of the Quetzal
(N. Y., 1887), and another by Bandelier in the Am. Antiq. Soc.
Proc., n. s., vol. i. p. 101. The references in Brasseur’s Hist.
Nations Civilisées, and in Bancroft’s Native Races, vol. v., will be a
ready means for collating the early sources.

[924] Scherzer and Brasseur are somewhat at variance here.

[925] “There are some coincidences between the Old Testament

and the Quiché MS. which are certainly startling.” Müller’s Chips,
i. 328.
[926] Wanderungen durch die mittel-Amerikanischen Freistaaten
(Braunschweig, 1857—an English translation, London, 1857).

[927] Leclerc, no. 1305.

[928] H. H. Bancroft, Nat. Races, ii. 115; iii., ch. 2, and v. 170, 547,
gives a convenient condensation of the book, and says that
Müller misconceives in some parts of his summary, and that
Baldwin in his Ancient America, p. 191, follows Müller. Helps,
Spanish Conquest, iv. App., gives a brief synopsis,—the first one
done in English.

[929] Max Müller dissents from this. Chips, i. 326. Müller reminds
us, if we are suspicious of the disjointed manner of what has
come down to us as the Popul Vuh, that “consecutive history is
altogether a modern idea, of which few only of the ancient
nations had any conception. If we had the exact words of the
Popul Vuh, we should probably find no more history there than
we find in the Quiché MS. as it now stands.”

[930] Cf. Aborig. Amer. Authors, p. 33.

[931] The names of the gods in the Kiché Myths of Central America
(Philad., 1881), from the Proc. Amer. Philos. Soc. He gives his
reasons (p. 4) for the spelling Kiché.

[932] Cf. Am. Antiq. Soc. Proc., n. s., vol. i. 109; and his paper, “On
the Sources of the Aboriginal Hist. of Spanish America,” in the
Am. Asso. Adv. Sci. Proc., xxvii. 328 (Aug., 1878). In the Peabody
Mus. Eleventh Report, p. 391, he says of it that “it appears to be
for the first chapters an evident fabrication, or at least
accommodation of Indian mythology to Christian notions,—a
pious fraud; but the bulk is an equally evident collection of
original traditions of the Indians of Guatemala, and as such the
most valuable work for the aboriginal history and ethnology of
Central America.”

[933] Hist. Nat. Civ., i. 47. S’il existe des sources de l’histoire
primitive du Méxique dans les monuments égyptiens et de
l’histoire primitive de l’ancien monde dans les monuments
Américains? (1864), which is an extract from his Landa’s Relation.
Cf. Bollaert, in the Royal Soc. of Lit. Trans., 1863. Brasseur (Bib.
Mex.-Guat., p. 45; Pinart, no. 231) also speaks of another Quiché
document, of which his MS. copy is entitled Titulo de los Señores
 de Totonicapan, escrito en lengua Quiché, el año de 1554, y
traducido al Castellano el año de 1834, por el Padre Dionisio José
Chonay, indígena, which tells the story of the Quiché race
somewhat differently from the Popul Vuh.

[934] See Vol. II. p. 419.

[935] It stands in Brasseur’s Bib. Mex.-Guat., p. 13, as Memorial de

Tecpan-Atitlan (Solola), histoire des deux familles royales du
royaume des Cakchiquels d’Iximché ou Guatémala, rédigé en
langue Cakchiquèle par le prince Don Francisco Ernantez Arana-
Xahila, des rois Ahpozotziles, where Brasseur speaks of it as
analogous to the Popul Vuh, but with numerous and remarkable
variations. The MS. remained in the keeping of Xahila till 1562,
when Francisco Gebuta Queh received it and continued it (Pinart
Catalogue, no. 35).

[936] See Vol. II. 419; Bancroft, Nat. Races, v. 564; Bandelier in
Am. Antiq. Soc. Proc., i. 105. Bandelier (Peabody Mus. Repts., ii.
391) says that it is now acknowledged that the Recordacion
florida of Fuentes y Guzman is “full of exaggerations and
misstatements.” Brasseur (Bib. Mex.-Guat., pp. 65, 87), in
speaking of Fuentes’ Noticia histórica de los indios de Guatemala
(of which manuscript he had a copy), says that he had access to
a great number of native documents, but profited little by them,
either because he could not read them, or his translators
deceived him. Brasseur adds that Fuentes’ account of the Quiché
rulers is “un mauvais roman qui n’a pas le sens commun.” This
last is a manuscript used by Domingo Juarros in his Compendio
de la historia de la ciudad de Guatemala (Guatemala, 1808-1818,
in two vols.—become rare), but reprinted in the Museo
Guatemalteco, 1857. The English translation, by John Baily, a
merchant living in Guatemala, was published as a Statistical and
Commercial History of Guatemala (Lond., 1823). Cf. Vol. II. p.
419. Francisco Vazquez depended largely on native writers in his
Crónica de la Provincia de Guatemala (Guatemala, 1714-16). (See
Vol. II. p. 419.)

[937] See note in Bancroft, iii. 451.

[938] Vol. II. 419. Helps (iii. 300), speaking of Remesal, says: “He
had access to the archives of Guatemala early in the seventeenth
century, and he is one of those excellent writers so dear to the
students of history, who is not prone to declamation, or rhetoric,
 or picturesque writing, but indulges us largely by the introduction
everywhere of most important historical documents, copied boldly
into the text.”

[939] Vol. II. 419.

[940] Vol. II. 417.

[941] E. G. Squier printed in 1860 (see Vol. II. p. vii.) Diego Garcia
de Palacio’s Carta dirigida al Rey de España, año 1576, under the
English title of Description of the ancient Provinces of
Guazacupan, Izalco, Cuscatlan, and Chiquimula in Guatemala,
which is also included in Pacheco’s Coleccion, vol. vi. Bandelier
refers to Estevan Aviles’ Historia de Guatemala desde los tiempos
de los Indios (Guatemala, 1663). A good reputation belongs to a
modern work, Francisco de Paula Garcia Pelaez’s Memorias para
la Historia del antiguo reyno de Guatemala (Guatemala, 1851-53,
in three vols.).

[942] For details follow the references in Brasseur’s Nat. Civil.;

Bancroft’s Nat. Races; Stephens’s Nicaragua, ii. 305, etc. See the
introd. of Brinton’s Güegüence (Philad., 1883), for the Nahuas
and Mangues of Nicaragua.

[943] Leclerc, no. 1070. Bancroft summarized the history of these

ancient peoples in his vol. ii. ch. 2, and goes into detail in his vol.
v.

[944] He condenses the early Mexican history in his Mexico, i. ch. 7.

There are recent condensed narratives, in which avail has been
had of the latest developments, in Baldwin’s Ancient America, ch.
4, and Short’s North Americans of Antiquity.

[945] Mrs. Alice D. Le Plongeon has printed various summarized

popular papers, like the “Conquest of the Mayas,” in the Mag.
Amer. Hist., April and June, 1888.

[946] A list of Squier’s published writings was appended to the

Catalogue of Squier’s Library, prepared by Joseph Sabin (N. Y.,
1876), as sold at that time. By this it appears that his earliest
study of these subjects was a review of Buxton’s Migrations of
the Ancient Mexicans, read before the London Ethnolog. Soc.,
and printed in 1848 in the Edinb. New Philosoph. Mag., vol. xlvi.
His first considerable contribution was his Travels in Cent.
 America, particularly in Nicaragua, with a description of its
aboriginal monuments (London and N. Y., 1852-53). He
supplemented this by some popular papers in Harper’s Mag.,
1854, 1855. (Cf. Hist. Mag., iv. 65; Putnam’s Mag., xii. 549.) A
year or two later he communicated papers on “Les Indiens
Guatusos du Nicaragua,” and “Les indiens Xicaques du Honduras,”
to the Nouvelles Annales des Voyages (1856, 1858), and “A Visit
to the Guajiquero Indians” to Harper’s Mag., 1859. In 1860,
Squier projected the publication of a Collection of documents, but
only a letter (1576) of Palacio was printed (Icazbalceta, Bibl.
Mex., i. p. 326). He had intended to make the series more correct
and with fewer omissions than Ternaux had allowed himself. His
material, then the result of ten years’ gathering, had been largely
secured through the instrumentality of Buckingham Smith. (See
Vol. II. p. vii.)

[947] “Art of war and mode of warfare of the Ancient Mexicans”

(Peabody Mus. Rept., no. x.).
“Distribution and tenure of lands, and the customs with
respect to inheritance among the ancient Mexicans” (Ibid. no.
xi.).
“Special organizations and mode of government of the ancient
Mexicans” (Ibid. no. xii.).
These papers reveal much thorough study of the earlier writers
on the general condition of the ancient people of Mexico, and the
student finds much help in their full references. It was this
manifestation of his learning that led to his appointment by the
Archæological Institute,—the fruit of his labor in their behalf
appearing in his Report of an Archæological Tour in Mexico, 1881,
which constitutes the second volume (1884) of the Papers of that
body. In his third section he enlarges upon the condition of
Mexico at the time of the Conquest. His explorations covered the
region from Tampico to Mexico city.

[948] Library of Aboriginal American Literature, (Philadelphia.)

[949] James H. McCulloh, an officer of the U. S. army, published

Researches on America (Balt., 1816), expanded later into
Researches, philosophical and antiquarian, concerning the
original History of America (Baltimore, 1829). His fifth and sixth
parts concern the “Institutions of the Mexican Empire,” and “The
nations inhabiting Guatemala” (Field, no. 987).
 G. F. Lyon’s Journal of a residence and tour in the Republic of
Mexico (Lond., 1826, 1828).
Brantz Mayer’s Mexico as it was and as it is, and his more
comprehensive Mexico, Aztec, Spanish and Republican (Hartford,
1853), which includes an essay on the ancient civilization. Mayer
had good opportunities while attached to the United States
legation in Mexico, but of course he wrote earlier than the later
developments (Field, no. 1038).
The distinguished English anthropologist, E. B. Tylor’s
Anahuac; or, Mexico and the Mexicans, ancient and modern
(London, 1861), is a readable rendering of the outlines of the
ancient history, and he describes such of the archæological
remains as fell in his way.
H. C. R. Becher’s Trip to Mexico (London, 1880) has an
appendix on the ancient races.
F. A. Ober’s Travels in Mexico (1884).

[950] The important papers are:—Tome I. Brasseur de Bourbourg.

Esquisses d’histoire, d’archéologie, d’ethnographie et de
linguistique. Gros. Renseignements sur les monuments anciens
situés dans les environs de Mexico.—Tome II. Br. de Bourbourg.
Rapport sur les ruines de Mayapan et d’Uxmal au Yucatan. Hay.
Renseignements sur Texcoco. Dolfus, Montserrat et Pavie.
Mémoires et notes géologiques.—Tome III. Doutrelaine. Rapports
sur les ruines de Mitla, sur la pierre de Tlalnepantla, sur un mss.
mexicain (avec fac-simile). Guillemin Tarayre. Rapport sur
l’exploration minéralogique des régions mexicaines. Siméon. Note
sur la numération des anciens Mexicains.

[951] He says the work is very rare. A copy given by him is in

Harvard College library. Bib. Mex.-Guat., p. 26.

[952] His Palenqué, at a later day, was published by the French

government (Quatre Lettres, avant-propos).

[953] Introduction of his Hist. Nations Civilisées.

[954] Tome I. xcii. et 440 pp. Les temps héroïques et l’histoire de

l’empire des Toltèques.—Tome II. 616 pp. L’histoire du Yucatan et
du Guatémala, avec celle de l’Anahuac durant le moyen âge
aztèque, jusqu’à la fondation de la royauté à Mexico.—Tome III.
692 pp. L’histoire des Etats du Michoacan et d’Oaxaca et de
l’empire de l’Anahuac jusqu’à l’arrivée des Espagnols. Astronomie,