Visit https://ebookultra.

com to download the full version and

explore more ebooks

Coefficient of Variation and Machine Learning

Applications 1st Edition K. Hima Bindu (Author)

_____ Click the link below to download _____

https://ebookultra.com/download/coefficient-of-
variation-and-machine-learning-applications-1st-
edition-k-hima-bindu-author/

Explore and download more ebooks at ebookultra.com

Here are some recommended products that might interest you.
You can download now and explore!

Domain Specific Computer Architectures for Emerging

Applications Machine Learning and Neural Networks 1st
Edition Chao Wang
https://ebookultra.com/download/domain-specific-computer-
architectures-for-emerging-applications-machine-learning-and-neural-
networks-1st-edition-chao-wang/
ebookultra.com

Bayesian reasoning and machine learning Barber D.

https://ebookultra.com/download/bayesian-reasoning-and-machine-
learning-barber-d/

ebookultra.com

Python Machine Learning By Example The Easiest Way to Get

into Machine Learning 1st Edition Yuxi (Hayden) Liu

https://ebookultra.com/download/python-machine-learning-by-example-
the-easiest-way-to-get-into-machine-learning-1st-edition-yuxi-hayden-
liu/
ebookultra.com

Kernel Methods and Machine Learning S. Y. Kung

https://ebookultra.com/download/kernel-methods-and-machine-learning-s-
y-kung/

ebookultra.com
A First Course in Machine Learning Chapman Hall CRC
Machine Learning Pattern Recognition 2nd Edition Simon
Rogers
https://ebookultra.com/download/a-first-course-in-machine-learning-
chapman-hall-crc-machine-learning-pattern-recognition-2nd-edition-
simon-rogers/
ebookultra.com

Python Machine Learning 1st Edition Wei-Meng Lee

https://ebookultra.com/download/python-machine-learning-1st-edition-
wei-meng-lee/

ebookultra.com

Statistical Machine Learning 1st Edition Richard M. Golden

https://ebookultra.com/download/statistical-machine-learning-1st-
edition-richard-m-golden/

ebookultra.com

Automated Machine Learning in Action 1st Edition Qingquan

Song

https://ebookultra.com/download/automated-machine-learning-in-
action-1st-edition-qingquan-song/

ebookultra.com

Implementation of Machine Learning Algorithms Using

Control Flow and Dataflow Paradigms 1st Edition
Milutinovi?
https://ebookultra.com/download/implementation-of-machine-learning-
algorithms-using-control-flow-and-dataflow-paradigms-1st-edition-
milutinovi/
ebookultra.com
Coefficient of Variation and Machine Learning
Applications 1st Edition K. Hima Bindu (Author) Digital
Instant Download
Author(s): K. Hima Bindu (Author); Raghava Morusupalli (Author); Nilanjan
Dey (Author); C. Raghavendra Rao (Author)
ISBN(s): 9781000752625, 1000752429
Edition: 1
File Details: PDF, 2.99 MB
Year: 2019
Language: english
Coefficient of Variation
and Machine Learning
­Applications
Coefficient of Variation
and Machine Learning
­Applications

K. Hima Bindu
Raghava M
Nilanjan Dey
C. Raghavendra Rao
CRC Press
Taylor & Francis Group
52 Vanderbilt Avenue,
New York, NY 10017

© 2020 by Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

International Standard Book Number-13: 978-0-367-27328-6 (Hardback)

This book contains information obtained from authentic and highly regarded sources.
Reasonable efforts 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. The 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, microfilming, 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, please access
www.copyright.com (http://www.copyright.com/) or contact the Copyright Clearance
Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-
for-profit organization that provides licenses and registration for a variety of users. For
organizations that have been granted a photocopy license by the CCC, a separate system of
payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered

trademarks, and are used only for identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com

and the CRC Press Web site at

http://www.crcpress.com
Contents

List of Figures, ix
List of Tables, xi
Preface, xiii
Authors, xvii

CHAPTER 1  Introduction to Coefficient of Variation 1

1.1 INTRODUCTION 1
1.2 COEFFICIENT OF VARIATION 3
1.3 NORMALIZATION 5
1.3.1 Coefficient of Variation of Normalized
Variable 5
1.3.2 Illustration 7
1.3.3 Random Variable with Probability Density
Function 8
1.3.3.1 Normal Distribution with Mean
μ(= 0) and Standard Deviation σ 8
1.3.3.2 Gamma Distribution with Mean
μ and Standard Deviation σ 8
1.3.4 Random Variable with Probability Mass
Function 8

v
vi  Contents

1.4 PROPERTIES OF COEFFICIENT OF

VARIATION 9
1.4.1 Properties of Mean 9
1.4.2 Properties of Standard Deviation 9
1.4.3 Properties of CV 9
1.4.4 Features Based on Coefficient of Variation 10
1.4.4.1 Influence of Translation and Scale
on Features of CV 11
1.4.5 CV of Mixture Distributions 13
1.4.5.1 Mixture of Normal Distributions 16
1.5 LIMITATIONS OF COEFFICIENT OF
VARIATION 22
1.6 CV INTERPRETATION 22
1.7 SUMMARY 23
1.8 EXERCISES 24

CHAPTER 2  CV Computational Strategies 27

2.1 INTRODUCTION 27
2.2 CV COMPUTATION OF POOLED DATA 28
2.2.1 Illustration 30
2.3 COMPARISON OF CV WITH ENTROPY
AND GINI INDEX 31
2.4 CV FOR CATEGORICAL VARIABLES 32
2.4.1 Table Lookup Method 33
2.4.2 Mapping Method 34
2.4.3 Zero Avoiding Calibration 35
2.5 CV COMPUTATION BY MAPREDUCE
STRATEGIES 36
2.6 SUMMARY 38
2.7 EXERCISES 38
 Contents  vii

CHAPTER 3  Image Representation 43

3.1 INTRODUCTION 43
3.2 CVIMAGE 44
3.2.1 CVImage Representation of Lena Image 45
3.3 CV FEATURE VECTOR 48
3.3.1 Demonstration 50
3.3.2 Hvalue Distribution Analysis 51
3.3.3 Ranking of CV Features 54
3.4 SUMMARY 55
3.5 EXERCISES 55

CHAPTER 4  Supervised Learning 57

4.1 INTRODUCTION 57
4.2 PREPROCESSING (DECISION ATTRIBUTE
CALIBRATION) 58
4.3 CONDITIONAL CV 59
4.4 CVGAIN (CV FOR ATTRIBUTE SELECTION) 61
4.4.1 Example 61
4.5 ATTRIBUTE ORDERING WITH CVGAIN 61
4.6 CVDT FOR CLASSIFICATION 63
4.6.1 CVDT Algorithm 64
4.6.2 CVDT Example 64
4.6.3 Using CVDT for Classification 67
4.7 CVDT FOR REGRESSION 68
4.8 CVDT FOR BIG DATA 71
4.8.1 Distributed CVDT Induction with
Horizontal Fragmentation 71
4.8.2 Distributed CVDT Induction with
Vertical Fragmentation 72
viii  Contents

4.9 FUZZY CVDT 73

4.10 SUMMARY 76
4.11 EXERCISES 76

CHAPTER 5  Applications 79
5.1 IMAGE CLUSTERING 79
5.2 IMAGE SEGMENTATION 82
5.3 FEATURE SELECTION 82
5.4 MOOD ANALYSIS 84
5.4.1 Bipolar Disorder 84
5.4.2 Twitter Mood Predicts the Stock Market 84
5.5 CV FOR OPTIMIZATION 85
5.6 HEALTH CARE 86
5.7 SOCIAL NETWORK 87
5.8 SUMMARY 87
5.9 EXERCISES 87

APPENDIX A, 89

REFERENCES, 121

INDEX, 125
List of Figures

Figure 1.1 Influence of translation and scale on mean 12

Figure 1.2 Influence of translation and scale on
standard deviation 12
Figure 1.3 Influence of translation and scale on CV 13
Figure 1.4 Influence of translation and scale on
normalized mean 14
Figure 1.5 Influence of translation and scale on
normalized standard deviation 14
Figure 1.6 Influence of translation and scale on
normalized CV 15
Figure 1.7 Mixture distributions example ( f1 has
μ = 4, σ = 1; f2 has μ = 6, σ = 2) 16
Figure 1.8 α Ratio 1:1 17
Figure 1.9 α Ratio 1:2 18
Figure 1.10 α Ratio 2:1 18
Figure 2.1 Binomial distributed variable 32
Figure 2.2 Normal distributed variable 33
Figure 2.3 Binomial distribution with zero avoidance 35
Figure 3.1 CVImage illustration: (a) Pixels values,
(b) CVImage demo, and (c) CVIF demo 46
Figure 3.2 CVImage: (a) Sample image, (b) CVImage
of sample image, and (c) CVImage with
CVT = 33% for edge representation 47

ix
x  List of Figures

Figure 3.3 CVImage of Lena: (a) Lena image and its

variants; (b) CVImage with CVT = 3, 9, 15,
21; (c) CVImage with CVT = 27, 33, 39, 45;
and (d) CVImage with CVT = 51, 57, 63, 69 48
Figure 3.4 Color image and histograms of its color
channels 49
Figure 3.5 SampleDB 51
Figure 4.1 CVDT for data in Table 4.2 67
Figure 4.2 Regression tree of AutoMPG data set 68
Figure 4.3 Variation of RMSE with alpha 70
Figure 4.4 RMSE of CVDT and CART 71
Figure 4.5 CVDT for fuzzy membership of Play = Yes 74
Figure 4.6 CVDT for fuzzy membership of Play = No 75
Figure 4.7 Fuzzy membership influence on RMSE and ESS 75
Figure 4.8 Two-dimensional data for classification 78
Figure 5.1 Elbow method on color images to determine
the optimal number of clusters with CV
Features. (a) Elbow method on 9 CV features.
(b) Elbow method on 3 CV features 80
Figure 5.2 Purity of clusters. (a) Purity with 9 CV
Features. (b) Purity with 3 CV features 81
Figure 5.3 Image segmentation with CVImage and
CVThreshold. (a) Brain tumor image;
(b) result of segmentation with
CVThreshold = 3; and (c) result of
segmentation with CVThreshold = 9 83
List of Tables

Table 1.1 Mixture of 2 normals with mixing α ratio 1:1 19

Table 1.2 Mixture of 2 normals with mixing α ratio 1:2 20
Table 1.3 Mixture of 2 normals with mixing α ratio 2:1 21
Table 2.1 Data for exercise 1 38
Table 2.2 Empirical distribution of data for exercise 1 38
Table 2.3 Super market transactions for exercise 3 40
Table 3.1 Image statistics 50
Table 3.2 Hash values 52
Table 3.3 Hash value distribution of sample DB 53
Table 3.4 Image databases 53
Table 3.5 Hvalue distribution database wise 54
Table 4.1 Train data for play tennis 58
Table 4.2 Play tennis data 60
Table 4.3 Decision table for Outlook = Sunny 66
Table 4.4 Decision table for Outlook = Overcast 66
Table 4.5 Decision table for Outlook = Rain 66
Table 4.6 Play tennis data with fuzzy memberships 74
Table 4.7 Decision system 76
Table 4.8 Employee data set 77
Table 4.9 Regression data set 78
Table A.1 CV features of Greenlake 91

xi
xii  List of Tables

Table A.2 CV features of Greenland in gray scale

(Hvalue = 1) 94
Table A.3 CV features of TrainDB 98
Table A.4 CV features of TestDB 109
Table A.5 CV features of Indonesia 114
Table A.6 CV features of SampleDB 116
Table A.7 CV features of Testimages 117
Table A.8 CV features of Images_Texture 119
Preface

Coefficient of Variation and Machine Learning

T HIS BOOK
Applications is an outcome of the experiences of the authors.
It is observed that several statistical measures like mean, standard
deviation, and moments are adopted as features for understanding
and representing systems. The coefficient of variation (CV), the
percentage relative variation to its mean, has received less atten-
tion in the domain of Data Mining and Machine Learning.
Coefficient of variation has been used for making decisions
in domains like Economics, Business Administration, and Social
Science. Coefficient of Variation indicates the constancy aspects
of the system (data). It is a potential candidate for consideration
as one of the features for representing a system/object/subject.
Coefficient of variation can also be used as an index to rank order
or compare the systems/objects/subjects in place of Gini Index
(measure of inequality), Entropy (measure of disorder), etc. Coef-
ficient of variation will support in designing computationally effi-
cient single pass algorithms due to its elegant algebraic properties.
The book is organized into five chapters. It is intended to pro-
vide introduction, computational strategies, representation, some
of the Machine Learning and Data Mining methodologies, and few
potential applications.
Chapter 1 is an introduction to Coefficient of Variation. The
reader can learn basic ideas, definitions and computation of CV
for a data vector and a random variable. It describes normaliza-
tion and transformation issues. Properties of CV, features that can
be derived from CV, possible interpretations based on CV and its

xiii
xiv  Preface

limitations are discussed in this chapter. This chapter sets stage for
subsequent chapters.
Chapter 2 presents the distributed computational strategies of
CV using pooled and Map-Reduce approaches. It also presents the
pre-processing method to be used for categorical data to enable
CV computations. This helps CV to be applicable for real-world
data used in Machine Learning.
Chapter 3 proposes using CV as an operator for producing
CVImage as a derived representation for an image. It presents
image binarization by tuning a threshold on CV. Using features
derived based on CV of the RGB colors, a CVFeature vector as
a knowledge representation mechanism is developed, which has
interesting applications.
Chapter 4 introduces supervised learning using CV. It defines
CVGain as an attribute selection measure of classification and
regression trees. The suitability of CVGain for distributed envi-
ronments and fuzzy decision systems is discussed.
Chapter 5 presents applications like image clustering and seg-
mentation by using CVImage features. It discusses CV-based fea-
ture ordering and feature selection, leading to a candidate order in
all Data Mining and Machine Learning approaches. Details of how
CV is potential to perform Mood analysis, provide solution for
optimization problem are outlined in this chapter. It ends with few
pointers to address health care and social network applications.
This book can be used as a concise handbook as well as a part
of courses in Data Mining, Machine Learning and Deep Learn-
ing. The authors feel that this book can help researchers in solving
Machine Learning applications with its decent coverage of illus-
trations and demonstrations.
Raghavendra Rao C. expresses his special thanks to
Mr. K. Omprakash, General Manager—Education Sales, Capri-
cot Technologies Private Limited for providing Maltab academic
license which is used in generating some of the results and figures.
 Preface  xv

Hima Bindu would like to acknowledge the Director—Prof.

C. S. P. Rao, the In-charge Registrar—Prof. G. Amba Prasad Rao
and faculty members of Computer Science and Engineering
department at NIT Andhra Pradesh for their encouragement and
cooperation.
The authors thank Mr. M. Ramakanth, BSNL, for valuable
inputs related to interval estimations of CV. The efforts of
Mr. Adithya Chillarige and Dr. Tilottama Goswami are acknowl-
edged for sharing the images and image databases.
Authors

Dr. K. Hima Bindu is an alumnus of

NIT—Warangal, JNTU—Hyderabad,
and University of Hyderabad. She
has a decade of experience in Data
Mining and Machine Learning. She
is currently serving as an assis-
tant professor in National Insti-
tute of Technology, Tadepalligudem,
India. Her research interests include
Data Mining, Machine Learning, Big
Data, and Educational Data Mining.

Dr. Raghava M obtained his MTech

from Mysore University, Karnataka,
India, during 2003. He received his
PhD from University of Hyderabad,
Telangana, India. He started his
engineering teaching career in CVR
College of Engineering, Hyderabad,
India, during 2003 and successfully
handled various courses related to
Systems Engineering, Neural Net-
works, Data Engineering, and Linux Internals. Currently he is
serving as a professor in CVR College of Engineering. His areas of
interests include Computer Vision, Regularization Theory, Sparse
Representations, and Graphical Models.

xvii
xviii  Authors

Nilanjan Dey is an assistant profes-

sor in Department of Information
Technology at Techno India Col-
lege of Technology (under Techno
India Group), Kolkata, India. He
is a visiting fellow of University of
Reading, London, UK, and visit-
ing professor of Duy Tan Univer-
sity, Vietnam. He was an honorary
visiting scientist at Global Biomed-
ical Technologies Inc., California
(2012–2015). He is a research scientist of Laboratory of Applied
Mathematical Modeling in Human Physiology, Territorial Orga-
nization of Scientific and Engineering Unions, Bulgaria, and asso-
ciate researcher of Laboratoire RIADI, University of Manouba,
Tunisia. He is a scientific member of Politécnica of Porto. He was
awarded his PhD from Jadavpur University, West Bengal, India, in
2015. In addition, recently he was awarded as one among the top
10 most published academics in the field of Computer Science in
India (2015–2017) during “Faculty Research Awards” organized
by Careers 360 at New Delhi, India. Before he joined Techno
India, he was an assistant professor of JIS College of Engineering
and Bengal College of Engineering and Technology, West Bengal,
India.
He has authored/edited more than 50 books with Elsevier,
Wiley, CRC Press, and Springer, and published more than 300
papers. His h-index is 32 with more than 4900 citations. He is the
editor-in-chief of International Journal of Ambient Computing and
Intelligence (IJACI, IGI Global, UK, Scopus), International Journal
of Rough Sets and Data Analysis (IGI Global, US, DBLP, ACM dl),
co-editor-in-chief of International Journal of Synthetic Emotions
(IGI Global, US, DBLP, ACM dl), and International Journal of Nat-
ural Computing Research (IGI Global, US, DBLP, ACM dl). He is
the series co-editor of Springer Tracts in Nature-Inspired Com-
puting (STNIC), Springer; series co-editor of Advances in Ubiq-
 Authors  xix

uitous Sensing Applications for Healthcare (AUSAH), Elsevier;

series editor of Computational Intelligence in Engineering Prob-
lem Solving and Intelligent Signal processing and data analysis,
CRC Press (FOCUS/Brief Series); and Advances in Geospatial
Technologies (AGT) Book Series (IGI Global), US. He serves as an
editorial board member of several international journals, includ-
ing International Journal of Image Mining (IJIM), Inderscience,
associated editor of IEEE Access (SCI-Indexed), and International
Journal of Information Technology, Springer.
His main research interests include Medical Imaging, Machine
learning, Computer Aided Diagnosis as well as Data Mining. He
has been on program committees of over 50 international con-
ferences, a workshop organizer of 5 workshops, and acted as a
program co-chair and/or advisory chair of more than 10 inter-
national conferences. He has given more than 40 invited lec-
tures in 10 countries, including many invited plenary/keynote
talks at the international conferences such as ITITS2017 (China),
TIMEC2017 (Egypt) and SOFA2018 (Romania), BioCom2018
(UK), etc.

Prof. C. Raghavendra Rao com-

pleted his BSc and MSc in Statistics
from Andhra University, Andhra
Pradesh, India, and Osmania Uni-
versity, Telangana, India, respec-
tively, and PhD in Statistics and
MTech (CS & Engineering) from
Osmania University.
He started his carrier as a lec-
turer in Statistics at Osmania Uni-
versity in 1984. Since 1986, he is
working in the School of Mathematics and Computer/Information
Sciences, University of Hyderabad, India. Presently he is a senior
professor in the School of Computer and Information Sciences,
University of Hyderabad.
xx  Authors

His current research interests are Simulation and Modeling,

Rough Sets, and Knowledge Discovery.
Dr. Rao is a member of the Operation Research Society of India,
Indian Mathematical Society, International Association of Engi-
neers, Society for development of Statistics, Andhra Pradesh Soci-
ety for Mathematical Sciences, Indian Society for Probability and
Statistics, Society for High Energy Materials, International Rough
Set Society, Indian Society for Rough Sets, and ACM. He was the
founder secretory of Indian Society for Rough Sets and also a
fellow of The Institution of Electronics and Telecommunication
Engineers, Society for Sciences, and Andhra Pradesh Akademi of
Science. He is a senior elected member for International Rough Set
Society 2016.
He has over 175 journals and conference proceeding papers to
his credit.
Books in his credit are:

• Evaluation of Total Literacy Campaigns (1995) co-authored

with Prof. Bh. Krishnamurti and Dr. I. Ramabrahmam, pub-
lished by Booklinks Corporation, Hyderabad.
• Multi-disciplinary Trends in Artificial Intelligence: 8th Inter-
national Workshop, MIWAI 2014, Bangalore, India, Decem-
ber 8–10, 2014, Proceedings, 8875, co-editors Murty,
M. Narasimha; He, Xiangjian; Weng, Paul published by
Springer.
• Strategic Polysemantic search for Web Mining (2013) co-
authored by Dr. K. Hima Bindu, Publisher Lambert Aca-
demic Publishing.

Restricted Report
“Integrated Low Fidelity Model for Fuel Air Explosive”
(2013)—A Restricted Report from ACRHEM to HEMRL (DRDO),
Pune submitted co-authored with Prof. Arun Agarwal, Prof.
Rajeev Wankar, Dr. Apparao, and Smt. Vijay Lakshmi.
 Authors  xxi

Dr. Rao contributed enormously in the project of National

Interest, few to mention are:

1. Mathematical model-based Control system for swing out

operation for the liquid nitrogen feeding boom for rocket
launching pad in Sree Hari Kota (T.A. Hydraulics Ltd.).

2. Design of Optimal driving profile generation for diesel loco-

motives for Indian Railways (Consultancy project for Medha
Servo Control System).

3. Mathematical modeling for 6 degree of Motion platform for

building driver training simulator.
4. Mathematical modeling fuel air explosive system for Defense
Research Development Organization.

5. Optimal design configuration methodologies for aerospace

applications with IIT Bombay for Defence Research and
Development Laboratory.

6. Modeling and Simulation of scene of offence Crime Nos. 42

& 43/2015 for Special Investigation Team.

7. Real-time data study and analysis, Medha Servo Drivers Pvt.

Ltd., 2017–2018.

8. Adding intelligence to second generation ATGM, ZEN

Technologies Pvt. Ltd. (2018–2019).
 CHAPTER 1

Introduction to
Coefficient of
Variation

1.1 INTRODUCTION
The growth in communication, Internet, and computing tech-
nologies is persistently posing problems to the researchers. The
data getting collected are becoming unmanageable, and develop-
ments in this direction led to the paradigm of research known
as “Data Science” or “Big Data.” The experience of automation
for past two decades is quite encouraging and given a hope, as
Machine Learning methods with computational intelligence sys-
tems are producing promising results. Hence, we are going to
experience higher level of comfort, integrity, quality of life, and
so on.
The sophisticated systems getting built with the help of these
developments resulted in new life style, namely, e-governess,
smart cities, healthcare, paperless transactions, mobility-free
transactions, and so on. The smartness in performance (opera-
tions) of these systems is due to consideration of all variations in

1
2  CV and Machine Learning Applications

the design of the systems such that the recommended (adopted)

decisions will be apt for the given context of the target state and
environment.
It is time to relook at what the scientific methodologies can
contribute to this futuristic systems architecture. The decision-
making processes in general are associated with matching patterns
that are already registered and then adopt to more or less the sim-
ilar decision to current situation. The matching process is highly
time consuming. To speed up this matching process, several meth-
ods are being designed and constructed. Most of these tools are
based on data analysis.
Expert systems (decision support system and knowledge
support system) provided tools for decision-making by encap-
sulating rules and built appropriate inference engines more in
medical domain (Mycin [1] is an example). The tractability lim-
itations of building knowledge and inference engines came in the
way of growth in Artificial Intelligence and Expert Systems. Tech-
nological growth in databases, Data Mining, and pattern recogni-
tion tools gave a new avenue to perform knowledge discovery in
databases. This led rule generation and knowledge representation
in an automated manner. With an assumption that data instance
is an apt representative for a particular domain, the knowledge
and rules discovered can be treated as domain knowledge, which
can be utilized for building inference engines. Inference engines
are mainly for making decisions. Thus, a recommender system
is analogous to the decision support system. As this process
is around data instance, these methods are called data-centric
methods.
One needs to take a decision among alternatives looking at
adaptability, performance, and aptness in a given system with a
fixed set of objectives, which is nothing but study of variations
in performance of various strategies. A decision maker as one of
the players, is trying to adopt a strategy that maximizes the profit
and minimizes the loss by modeling the above as a game theory
problem.
 Introduction to Coefficient of Variation  3

An apt ranking process of the strategies is in demand for

building recommendation system. This ranking process can be
addressed by various methods that are based on information
theory and variational methods. The statistical measures like
(information) Entropy, Gini Index, Variance, and Coefficient of
Variation (CV) are in use for decision-making. Several Machine
Learning contributions are mostly based on Entropy and Gini
Index. The variational methods like chi square and variance are
also adopted for rule discovery, testing and validation, and feature
ordering and feature selection [2, 3].
Although CV is one of the popular methods used for decision-
making by managers and risk analysts, it is yet to attract the atten-
tion of Data Mining and pattern recognition experts.
The variance and mean are the commonly used statistical mea-
sures to describe a system under study. Variance contributes the
variation w.r.t. the location, and the mean is a location. These mea-
sures are highly sensitive to the units (mathematically, scale and
translation). CV by definition is percentage of relative variation to
its mean. It obeys some of the index numbers properties [4].
The Section 1.2 provides Coefficient of Variation definition and
its computation. Section 1.3 discusses CV for normalized data.
Sections 1.4 and 1.5 presents properties of CV and its limitations.
Interpretation aspects of CV are provided in Section 1.6.

1.2 COEFFICIENT OF VARIATION

Coefficient of variation [5] abbreviated as CV is a statistical mea-
sure defined as percentage relative variation to its mean of a given
data (population) given by equation 1.1

σ
CV = ∗ 100 (1.1)
μ
where σ and μ (μ = 0) are the standard deviation and the mean
of the data (population), respectively. Estimate of CV and interval
estimate of CV can be found in [6].
4  CV and Machine Learning Applications

The absolute coefficient of variation (ACV) is defined as the

percentage of relative variation to its absolute mean of a given data
(population) using the following equation 1.2.
σ
ACV = ∗ 100 (1.2)
|μ|

where | μ | indicates absolute value of μ.

When X is a random variable, which can be discrete or contin-
uous, with probability mass function p(x) and probability density
function f (x), respectively.
Algorithm 1 returns the mean μ, standard deviation σ , coeffi-
cient of variation CV, and absolute coefficient of variation ACV
for a given input data (real valued) vector X.
Algorithm 2 returns the mean μ, standard deviation σ , coeffi-
cient of variation CV, and absolute coefficient of variation ACV
for a given random variable X.

Algorithm 1 Computation of Coefficient of Variation for a

Given Data
Require: X  data vector
Ensure: μ, σ , CV, and ACV  mean, standard deviation,
coefficient of variation, and absolute coefficient of variation
1: S ← 0  Sum
2: SS ← 0  Sum of Squares
3: n ← length(X )
4: for i ← 1 . . . n do
5: S ← S + X (i)
6: SS ← SS + X (i) ∗ X (i)
7: end for
8: μ ← n  mean
S

9: v ← n − μ2  variance
SS
√
10: σ ← v  standard deviation
σ
11: CV ← μ ∗ 100  CV
σ
12: ACV ← |μ| ∗ 100  ACV
 Introduction to Coefficient of Variation  5

Algorithm 2 Computation of Coefficient of Variation for Given

Probability Model (Population)
Require: Probability model of a random variable X  PDF or
PMF
Ensure: μ, σ , CV, and ACV  mean, standard deviation,
coefficient of variation, and absolute coefficient of variation
1: μ ← E(X )  expected value of X
2: V (X ) ← E(X ) − μ
2 2
 E(X ) is expectation
2 2
σ
 of X
3: CV ← μ ∗ 100  σ = (V (X ))
σ
4: ACV ← |μ| ∗ 100

1.3 NORMALIZATION
Most of the data (Y) accessible for analytics is a transformed form
of the original (X). In general, this transformation is normalized
transformation (linear) such that the transformed values will be
between 0 and 1. The abstract of the same is discussed below and
hence the interrelationship of X and Y.
Let X be given bounded real (random) variable with a and b be
the lower and upper sharp bounds. If a ≤ X ≤ b, then r = b − a.
r can also be computed as the least difference of arbitrary upper
bound and arbitrary lower bound.
Let Y be the normalized variable of X. The Y is obtained by the
below transformation:
X−a X−a
Y= = .
b−a r
Application of this transformation is known as normalization.
Algorithm 3 presents the normalization procedure.

1.3.1 Coefficient of Variation of Normalized Variable

Normalization is nothing but a linear transformation

1 −a
Y= X+
( b − a) ( b − a)
6  CV and Machine Learning Applications

Algorithm 3 Normalization
Require: X  X is vector of real numbers or random variable
with PDF f (.) or PMF p(.)
Ensure: Y, a, b, r
1: a ← min(X )  min(X) if X is data vector, else minimum of
the nontrivial support of f (.) or p(.)
2: b ← max(X )  max(X) if X is data vector, else maximum of
the nontrivial support of f (.) or p(.)
3: r ← b − a
X −a
4: Y ← r

1 −a
E (Y ) = E (X ) +
( b − a) ( b − a)

(E(Y ) > 0 for non-trivial distributions)

1
V (Y ) = V (X )
( b − a) 2
Hence, coefficient of variation of Y is
    
V (Y ) V (X ) 1 −a
CV (Y ) = = E (X ) +
E (Y ) ( b − a) ( b − a) ( b − a)

= V (X ) ( E ( X ) − a)

CV (Y ) > 0 for nontrivial distributions.

√
 V (X )
V (X ) E (X )
CV (Y ) = =
E (X ) − a 1 − E(aX)

CV (X ) ACV (X )
= =
1 − E (X )
a
D

where
1 − a/E(X ) when E(X ) > 0
D=
−1 + a/E(X ) when E(X ) < 0
 Introduction to Coefficient of Variation  7

Theorem 1: CV (Y ) is a non-negative.
CV (X )
Proof : CV (Y ) = 1− E(aX)
Since a ≤ X ≤ b, a is always smaller than E(X ) (for nontrivial
distributions with support [a, b]).
Hence, E(aX) < 1 if E(X ) > 0 and E(aX) > 1 if E(X ) < 0.
If E(X ) < 0, then CV (X ) < 0 and 1 − E(aX) < 0; hence ratio of
CV (X )
these two quantities is positive. That is, 1− E(aX) > 0.
Similarly, if E(X ) > 0, then CV (X ) > 0 and 1− E(aX) > 0; hence,
CV (X )
the ratio of these two quantities is positive. That is, 1− E(aX) > 0.
Hence, CV (Y ) is always positive.

1.3.2 Illustration
Let X be data vector of size 25, having the values:

[ 0 1 -2 0 -5 3 -5 -4 1 -2 1 -3 -4 -6 0 -2 -2 0 -6 -3 1 0 -5 -5 -6 ]
a = min(X ) = −6
b = max(X ) = 3
r=9
Y = (X + 6)/9
The normalized values are
0.67 0.78 0.44 0.67 0.11 1 0.11 0.22 0.78 0.44 0.78 0.33
0.22 0 0.67 0.44 0.44 0.67 0 0.33
E(X ) = −2.12 (note that E(X ) < 0)
V (X ) = 6.99
CV (X ) = −124.67
ACV (X ) = 124.67
a/E(X ) = 2.83(> 1)
E(Y ) = 0.43
V (Y ) = 0.09
CV (Y ) = 68.12
CV (X ) −124.67 −124.67
CV (Y ) = 1− E(aX) = 1−2.83 = −1.83 = 68.12
ACV (X ) 124.67 124.67
CV (Y ) = −1+ E(aX) = −1+2.83 = 1.83 = 68.12
8  CV and Machine Learning Applications

1.3.3 Random Variable with Probability Density Function

Let X be random variable with non trivial support [a, b], which
satisfies the required coverage probability β = Pr(a ≤ X ≤ b) =
b
a
f (x)dx, which is close to 1. For a choosen β , various pairs a,b
are possible. Among these possible pairs, pick up a pair with min-
imum b − a value.

1.3.3.1 Normal Distribution with Mean μ(= 0) and Standard

Deviation σ
CV (X ) = σμ ∗ 100
One can consider a = μ − kσ and b = μ + kσ . Here the
choice of k depends on the pre-defined coverage probability β =
Pr(−k ≤ Z ≤ k), where Z is a standard normal random variable.
r = 2kσ
Y = (x − (μ − kσ)) /(2kσ)
(X ) CV (X )
CV (Y ) = CVμ−
1− kσ = 1−1+k σ = k
100
μ μ

This CV is implicitly a function of β and independent of

parental characteristics.

1.3.3.2 Gamma Distribution with Mean μ and Standard

Deviation σ
One can consider a = 0 and b = μ + kσ , with the choice of
k depending on the coverage probability β = Pr(0 ≤ X ≤ b).
Then, r = μ + kσ , Y = X /(μ + kσ) and CV (Y ) = CV (X ).
Which means that CV of transformed variable is same as that of
the original variable. This result is obvious in normalization when
a = 0, and CV of normalized variable is same as CV of original
variable. Statistical feature CV is preserved intact in this type of
normalization [5, 7].

1.3.4 Random Variable with Probability Mass Function

Let X be random variable with non trivial support [a, b], which
satisfies the required coverage probability β = Pr(a ≤ X ≤ b) =
x =b
x=a Pr (X = x) and is preferably has shorter length interval.
 Introduction to Coefficient of Variation  9

In case of Poisson distribution with rate parameter λ, CV =

√1
λ
as the mean = variance = λ. The lower bound of the Poisson
variate is zero, hence a = 0. √
As a = 0 and one can consider b = λ + k λ, the choice of k
depends on √ β = Pr(0 ≤ X ≤ b). Hence,
√ the coverage probability
r = λ + k λ and Y = X /(λ + k λ), therefore CV (Y ) = CV (X ).
That is, CV of transformed variable is same as that of original vari-
able. As a = 0, it is similar to that in the case of Gamma distribu-
tion. Please note that this identity will not be true for truncated
(zero avoiding) or shifted distributions.

1.4 PROPERTIES OF COEFFICIENT OF VARIATION

1.4.1 Properties of Mean
Let Y = aX + b be a linear transformation of X, which indicates
the change of scale by a and then origin by b. Then, the mean of
Y is a × mean(X ) + b. That is, μY = a ∗ μX + b, where μY and
μX are means of Y and X, respectively. Hence, the transformation
applied on the feature X is to be applied on its mean to get mean
of transformed variable Y.

1.4.2 Properties of Standard Deviation

Let Y = aX + b be a linear transformation of X, which indicates
the change of scale by a and origin by b. Then, σY = a ∗ σX , where
σY and σX are standard deviations of Y and X, respectively. Hence,
standard deviation is insensitive with respective to the change of
origin.

1.4.3 Properties of CV
Let Y = aX + b be a linear transformation of X, which indicates
the change of scale a and then origin b.
a∗σX
Then, the CVY = a∗μ X +b
× 100 = aa∗+CVb X , where CVY and CVX
μX

are coefficients of Y and X, respectively, and μX and σX are mean

and standard deviation of X, respectively.
When b = 0, that is, when there is only scale transformation
(no change of origin), then CVY = CVX .
10  CV and Machine Learning Applications

Coefficient of variation is positive when mean is positive and

negative when mean is negative.
While absolute coefficient of variation is positive irrespective of
mean being positive or negative.
Absolute coefficient of variation will be the same as coefficient
of variation when the mean is positive.
Coefficient of variation is not definable when mean is zero.
In general, one can apply translation and then scaling to the
data, then the transformation will be YT = a(X − b), then
a ∗ σX σX
CVYT = × 100 = × 100
a ∗ μX − a ∗ b μX − b

Hence,
CVX
CVYT = (1.3)
1 − μbX
Equation 1.3 indicates that CV is invariant with respect to the
change of scale a but variant with respect to location (transla-
tion) b. CV approaches ∞ (unbounded) as b approaches mean
μX (assume μX > 0) from the left. The relative translation ratio
ρ = μbX need to be bounded below by −∞ and above by 1, that
is, −∞ < ρ < 1 to have non-negative CV. In normalization, the
parameter a (minimum of X) is equal to translation parameter b
in the above equation. While nontrivial support r = b − a is the
same as scaling parameter of the above equation.

1.4.4 Features Based on Coefficient of Variation

The aggregates or statistics like min, max, range, mean, standard
deviation, and moments can be calculated on features. These pro-
vide informative abstractions about a set of observations [8]. Using
thresholds as cuts is an important issue in Rough Sets [9] for dis-
cretization of continuous attributes. These cuts provide localized
information. A combination of the statistical aggregation and a
set of cuts will be useful in Machine Learning and deep learning.
This section considers a set of features named as “Features based
on CV” (CVFeatures), which are used in subsequent chapters for
 Introduction to Coefficient of Variation  11

demonstrating representation of objects as well as for learning. We

will discuss image representation using CV of a sliding window in
Chapter 3. Further, using threshold of 33%, we will develop flag,
leading to a bit vector that will be used as Hash value (Hvalue) in
Chapter 3. In Chapter 5, a Boolean decision system is generated to
perform mood analysis.
The data can be represented by using the following features:

1. min (minimum of the data)

2. max (maximum of the data)
3. S_mean (sign of mean of the data)
4. mean (magnitude of the mean of the data)
5. std (standard deviation of the data)
6. ACV (absolute coefficient of variation)
7. NCV (coefficient of variation of normalized data)
8. flag (ACV < CV_Threshold)
9. nflag (NCV < NCV_Threshold)

The features flag and nflag depend on the choice of threshold. A

required number of flags can be generated by choosing appropri-
ate set of thresholds.

1.4.4.1 Influence of Translation and Scale on Features of CV

Figures 1.1 through 1.3 depict mean, standard deviation, and
coefficient of variation and its variations/influence with respect
to change of translation and scale. The data considered for these
figures are uniformly distributed in the range [0, 1], translation
values are in the range [−10, 10], scale is in the range [0.1, 10]
with step size of 0.1. As given in Algorithm 3, normalized data
have been computed for each value of translation and scale for
the data used for Figures 1.1 through 1.3. One can notice the
monotonic behavior of mean after the transformation by both
of these from Figure 1.1. From Figure 1.2, it is evident that the
12  CV and Machine Learning Applications

FIGURE 1.1 Influence of translation and scale on mean.

FIGURE 1.2 Influence of translation and scale on standard deviation.

 Introduction to Coefficient of Variation  13

FIGURE 1.3 Influence of translation and scale on CV.

standard deviation is not effected due to translation and has lin-

ear effect with respect to scale. Figure 1.3 demonstrates that scale
has no effect on coefficient of variation, while translation influ-
ences exponentially (Note that the translation ratio considered is
−20.2 < ρ < 20.2). The mean, standard deviation, and coeffi-
cient of variation are computed for this normalized data and plot-
ted from Figures 1.4 through 1.6, respectively. As all these figures
resemble planes with directional cosines as (0, 0, 1), z-axes as per-
pendicular to the plane, indicating that translation and scale have
no effect for normalized data.

1.4.5 CV of Mixture Distributions

Consider k independent univariate random variables X1 , X2 , . . . ,Xk
with means μ1 , μ2 , . . . , μk and standard deviations σ1 , σ 2 , . . . , σk ,
respectively. The corresponding CVs of X1 , X2 , . . . , Xk are CV1 =
σ1 σ2 σk
μ1 ∗ 100, CV2 = μ2 ∗ 100, . . . , CVk = μk ∗ 100.
Let f (x) be the probability density function (PDF) of Y,
which is a mixture distribution of X1 , X2 , . . . , Xk with PDFs
f1 (x), f2 (x), . . . , fk (x), respectively. The respective mixing param-
eters are α1 , α2 , . . . , αk (where αi ≥ 0 and αi = 1).
14  CV and Machine Learning Applications

FIGURE 1.4 Influence of translation and scale on normalized mean.

FIGURE 1.5 Influence of translation and scale on normalized standard

deviation.
 Introduction to Coefficient of Variation  15

FIGURE 1.6 Influence of translation and scale on normalized CV.

Then, f (x) is computed as given in equation 1.4


k
f (x ) = α i fi ( x ) (1.4)
i= 1
CV of Y can be computed by using mean and standard devia-
tion of Xi , i = 1, 2, . . . k
Mean of Y is μY = α1 μ1 + α2 μ2 + · · · + αk μk
Expected value of Y 2 is E(Y 2 ) = α1 (σ12 + μ21 ) + α2 (σ22 + μ22 ) +
· · · + αk (σk2 + μ2k )
Variance of Y is V (Y ) = E(Y 2 ) − (E(Y )2 ) = {α1 (σ12 + μ21 ) +
α2 (σ22 + μ22 ) + · · · + αk (σk2 + μ√
k )} − (α1 μ1 + α2 μ2 + · · · + αk μk )
2 2

σY V (Y )
Now CV (Y ) = μY ∗ 100 = μY ∗ 100
CV(Y) =
 
  2  2
 μ μ
 α1 1
(CV12 +1002 ) + · · · + αk
k
(CVk2 +1002 ) −1002
μY μY
  
= ω1 (CV12 + 1002 ) + · · · + ωk CVk2 + 1002 − 1002 (1.5)

where ωi = αi ( μμYi )2 for i = 1, 2, . . . , k

16  CV and Machine Learning Applications

From equation 1.5, it is observed that CV (Y ) can be computed

by using αi , CVi , and μi of Xi , where i = 1, 2, . . . , k. Hence, it is
possible to compute CV (Y ) without revisiting rudimentary data.

1.4.5.1 Mixture of Normal Distributions

Let Y be the mixture of k univariate normal random variables
X1 , X2 , . . . , Xk with α1 , α2 , . . . , αk as mixing parameters. Then,
the probability density function (PDF) of Y be f (x) given as in
equation (1.6)

k
1 (xi −μi )2
f (x ) = αi √ e− 2 (1.6)
i= 1
σi 2π

Figure 1.7 shows an example of mixed distribution with α1 = 2

and α2 = 1. The base distributions are f1 and f2 . f1 has μ = 4 and
σ = 1, while f2 has μ = 6 and σ = 2.
Two normal distributions with means (μ1 , μ2 ), standard devi-
ations (σ1 , σ2 ) with mixing ratios (α1 , α2 ), where α1 + α2 = 1

FIGURE 1.7 Mixture distributions example (f1 has μ = 4, σ = 1; f2 has

μ = 6, σ = 2).
 Introduction to Coefficient of Variation  17

and α1 , α2 ≥ 0 are considered to demonstrate behavior of

mixture distribution. Consider the nine cases for (μ1 , μ2 ) =
{(4, 2), (4, 4), (4, 6)}; (σ1 , σ2 ) = {(1, 1), (1, 0.5), (1, 2)} for each
mixing ratio α1 = 1/2, 1/3, and 2/3 (α2 inferred by α1 + α2 = 1).
Figures 1.8 through 1.10 and Tables 1.1 through 1.3 are for α1 =
1/2, 1/3, and 2/3, respectively. Here, μ∗ , σ ∗ , and CV ∗ refer to
pooled values with the mixing ratios.
Each subplot of Figures 1.8 through 1.10 provides the PDF of
basic two normal distributions with mean and standard deviation
as given in the corresponding subplot for mixing α ratios of 1:1,
1:2, and 2:1, respectively. All these subplots follow the same leg-
end as in Figure 1.7. The bimodal behavior can be observed when
means are different relative to their respective standard deviations
in these plots.
Tables 1.1 through 1.3 for the respective figures provide sta-
tistical information (mean, standard deviation, and coefficient
of variation) of the base normal PDFs followed by the mixed

FIGURE 1.8 α Ratio 1:1.

18  CV and Machine Learning Applications

FIGURE 1.9 α Ratio 1:2.

FIGURE 1.10 α Ratio 2:1.

TABLE 1.1 Mixture of 2 Normals with Mixing α Ratio 1:1
S. No. α1 α2 μ1 μ2 σ1 σ2 CV 1 CV 2 μ* σ* CV *
1 0.50 0.50 4 2 1 1 25.00 50.00 3.00 1.41 47.14
2 0.50 0.50 4 2 1 0.5 25.00 25.00 3.00 1.27 42.49
3 0.50 0.50 4 2 1 2 25.00 100.00 3.00 1.87 62.36
4 0.50 0.50 4 4 1 1 25.00 25.00 4.00 1.00 25.00
5 0.50 0.50 4 4 1 0.5 25.00 12.50 4.00 0.79 19.76
6 0.50 0.50 4 4 1 2 25.00 50.00 4.00 1.58 39.53
7 0.50 0.50 4 6 1 1 25.00 16.67 5.00 1.41 28.28
8 0.50 0.50 4 6 1 0.5 25.00 8.33 5.00 1.27 25.50
9 0.50 0.50 4 6 1 2 25.00 33.33 5.00 1.87 37.42
Introduction to Coefficient of Variation  19
TABLE 1.2 Mixture of 2 Normals with Mixing α Ratio 1:2
S. No. α1 α2 μ1 μ2 σ1 σ2 CV 1 CV 2 μ* σ* CV *
10 0.33 0.67 4 2 1 1 25.00 50.00 2.67 1.37 51.54
11 0.33 0.67 4 2 1 0.5 25.00 25.00 2.67 1.18 44.19
12 0.33 0.67 4 2 1 2 25.00 100.00 2.67 1.97 73.95
13 0.33 0.67 4 4 1 1 25.00 25.00 4.00 1.00 25.00
14 0.33 0.67 4 4 1 0.5 25.00 12.50 4.00 0.71 17.68
15 0.33 0.67 4 4 1 2 25.00 50.00 4.00 1.73 43.30
16 0.33 0.67 4 6 1 1 25.00 16.67 5.33 1.37 25.77
17 0.33 0.67 4 6 1 0.5 25.00 8.33 5.33 1.18 22.10
20  CV and Machine Learning Applications

18 0.33 0.67 4 6 1 2 25.00 33.33 5.33 1.97 36.98

TABLE 1.3 Mixture of 2 Normals with Mixing α Ratio 2:1
S. No. α1 α2 μ1 μ2 σ1 σ2 CV 1 CV 2 μ* σ* CV *
19 0.67 0.33 4 2 1 1 25.00 50.00 3.33 1.37 41.23
20 0.67 0.33 4 2 1 0.5 25.00 25.00 3.33 1.28 38.41
21 0.67 0.33 4 2 1 2 25.00 100.00 3.33 1.70 50.99
22 0.67 0.33 4 4 1 1 25.00 25.00 4.00 1.00 25.00
23 0.67 0.33 4 4 1 0.5 25.00 12.50 4.00 0.87 21.65
24 0.67 0.33 4 4 1 2 25.00 50.00 4.00 1.41 35.36
25 0.67 0.33 4 6 1 1 25.00 16.67 4.67 1.37 29.45
26 0.67 0.33 4 6 1 0.5 25.00 8.33 4.67 1.28 27.43
27 0.67 0.33 4 6 1 2 25.00 33.33 4.67 1.70 36.42
Introduction to Coefficient of Variation  21
22  CV and Machine Learning Applications

distribution. The first column identifies the mixed distribution for

α1 and α2 mixing ratios, of base PDFs with means μ1 , μ2 , stan-
dard deviations σ1 , σ2 , and coefficient of variations CV1, CV2,
respectively. μ∗ , σ ∗ , and CV ∗ correspond to the resulting mixture
distribution.
Although the base distributions are symmetric, the mixed dis-
tribution f (.) needs not be symmetric. The CV of f (.) is depen-
dent on number of its peaks and geometric nature of the peaks.
The peaks of f (.) are influenced by base distribution’s components
and mixing ratios. In particular, the top middle subplots of each
Figures 1.8 through 1.10 have CV more than 33% for mixed dis-
tribution, though the base components are less than 33%. Rows
with identifiers 2, 11, and 20 in Tables 1.1 through 1.3 correspond
to these top middle subplots, and the CV of base distributions is
consistent in these rows (less than 33%). The geometric properties
of peaks and curvatures will help in understanding the feature that
has multimodal empirical density function (like relative frequen-
cies) with large CV.

1.5 LIMITATIONS OF COEFFICIENT OF VARIATION

Coefficient of variation is defined for numerical data (real random
variable). Coefficient of variation is not definable when mean is
zero, it will be unbounded when the mean is approaching zero.
It is recommended to convert all variables to ratio type by
appropriate transformation and calibrations [8]. To avoid the non
existence of CV, it is recommended to code in strictly positive
zone. Further, it is recommended to bring the range of normali-
zation to [1, 2]. So, add constant 1 to the normalization equations
used in Section 1.3. Coefficient of variation cannot be defined for
multivariate numerical data (multivariate real random variable).

1.6 CV INTERPRETATION
The coefficient of variation is a unit free index. ACV is always non-
negative. In most of the applications, the data will be non-negative,
hence CV is equivalent to ACV. In this section, CV is referred
Discovering Diverse Content Through
Random Scribd Documents
The Project Gutenberg eBook of The story of the
Rhinegold (Der Ring des Nibelungen) told for
young people
 This ebook is for the use of anyone anywhere in the United
States and most other parts of the world at no cost and with
almost no restrictions whatsoever. You may copy it, give it away
or re-use it under the terms of the Project Gutenberg License
included with this ebook or online at www.gutenberg.org. If you
are not located in the United States, you will have to check the
laws of the country where you are located before using this
eBook.

Title: The story of the Rhinegold (Der Ring des Nibelungen) told
for young people

Author: Anna Alice Chapin

Release date: May 5, 2024 [eBook #73548]

Language: English

Original publication: New York: Harper & Brothers Publishers,

1897

Credits: Charlene Taylor, Linda Cantoni, and the Online

Distributed Proofreading Team at https://www.pgdp.net
(This file was produced from images generously made
available by The Internet Archive/American Libraries.)

*** START OF THE PROJECT GUTENBERG EBOOK THE STORY

OF THE RHINEGOLD (DER RING DES NIBELUNGEN) TOLD FOR
YOUNG PEOPLE ***
 Transcriber's Notes
Anglicized variants of German character names have been retained as they are in the
original (e.g., Rhinegold for Rheingold, Logi for Loge, Friea for Freia). Obvious printer
errors have been corrected without note. Some full-page illustrations have been moved to
the nearest paragraph break for better readability.
To hear the music, click the [Listen] link, which will open an mp3 file. You can also
download a MusicXML file of the notation by clicking the [MusicXML] link. These links
may not work in e-book formats other than HTML. The music files are the music
transcriber’s interpretation of the printed notation and are placed in the public domain.

CONTENTS
 [Page 58

WOTAN AND BRÜNNHILDE

 THE
STORY OF THE
RHINEGOLD
(DER RING DES NIBELUNGEN)

Told for Young People

BY

ANNA ALICE CHAPIN

ILLUSTRATED

NEW YORK AND LONDON

HARPER & BROTHERS PUBLISHERS
1900

Copyright, 1897, by Harper & Brothers.

 All rights reserved.

TO

THE MASTER’S DAUGHTER

EVA WAGNER
WITH HEARTFELT GRATITUDE

FOR HER KINDNESS AND ENCOURAGEMENT

THIS BOOK

Is Dedicated
 PREFACE

The Story of the Rhinegold contains the four operas of Richard

Wagner’s “Nibelungen Ring,” arranged for young people. The
“Nibelungen Ring,” or “Nibelungen Cycle,” is built upon a colossal
foundation: a number of the great Teutonic myths, welded together
with the most masterly skill and consistency. It is evident that
Wagner, like William Morris and other writers, has taken from the
fragmentary mythological tales such material as would serve his
purpose, adapting such incidents as he chose and as he considered
appropriate to his work. But there are so many different versions of
these old stories that it is very difficult to trace Wagner’s plot to its
original birthplace. The various tales contained in the ancient sagas
are so seemingly contradictory that anything connectedly
authoritative appears impossible to trace. The one thing which
seems to remain the same in almost all versions of the stories,
ancient and modern, is the background of mythology, that great,
gloomy cycle of gods, with the ever-recurring note of Fate which
seems to have impressed all searchers in myths alike, and which
inspired Wagner when he formed his mystical, solemn Fate motif.
Odin, Wuotan, Wodin, or Wotan, according to the different names
given him in the old legends, is the central figure in the framework. If
I read the story aright, the Norns, or more properly Nornir, are next in
importance. They and their mother, the Vala, are the medium
through which the relentless something behind the gods made itself
felt in the world. The three sisters are named respectively Urðr,
Verðandi, and Skuld—freely translated Past, Present, and Future; or,
as they were once styled, as correctly perhaps, Was, Is, and Shall
Be. It is a question whether Erda and Urðr, the oldest Norn, might
not originally have been identical. Dr. Hueffer speaks of Erda as the
“Mother of Gods and Men,” but though “the Vala” is often found in
mythology, the name Erda is rarely mentioned, whereas the titles for
the three Norns seem to be unquestionably correct. The term Vala is
usually translated as Witch, or Witch-wife, but, though a Vala was
indeed a sorceress, she was a prophetess as well.
A step lower than the gods, yet gifted with supernatural power and
far removed from the characteristics of human beings, were the
dwarfs and the giants. The giants, we are told, were creatures
belonging properly to the Age of Stone, which explains the fact that
there were left but two representatives of the race at the time of the
Golden Age. The dwarfs come under the head of elves. They were
gifted with the utmost cleverness and skill. The giants were stupid
and clumsy, and, save for their superhuman strength and size,
entirely inferior to the small, sly dwarfs.
The world was strangely peopled in those days; many of the heroes
were demi-gods, that is, descended from some god or goddess, and
witches, dwarfs, and sorcerers mingled with human beings.
Many mortals, also, had magic power then. Otter, the son of Rodmar,
changed himself into the animal for which he was named, and while
in the shape of the otter he was caught and killed by three of the
gods who were wandering over the earth in disguise. Rodmar
demanded weregild,[A] and Loki, with a net, caught Andvari, a rich
and malignant dwarf, and commanded him to pay a ransom of gold
and gems, enough to cover the skin of the otter; for such was the
weregild demanded by Rodmar. Andvari, of necessity, gave the gold
for his own release, even adding a wonderful wealth-breeding Ring
to cover up a single hair in the skin which the rest of the treasures
had left unconcealed. The dwarf cursed the Ring, and the curse
attended it through all its manifold ways of magic, to the end of the
story.
[A] Weregild is almost untranslatable. It may mean payment, tax,
forfeit, or ransom.

Rodmar’s remaining sons, Fafnir and Regin, killed their father and
fought for the treasure. Fafnir obtained it, and, turning himself into a
monster-worm, went to Glistenheath (sometimes called Glittering
Hearth) to guard his wealth. Regin called upon Sigurd, a young hero,
to aid him, and, being a master-smith, forged for him a sharp sword
named Gram. Some versions give the forging of the sword to Sigurd,
but there are many sides to the story. The sword was sometimes
called Gram, and oftener Baldung, until Wagner gave it the more
expressive name of Nothung, or Needful. Prompted by Regin, Sigurd
slew the Dragon at Glistenheath, and, after tasting the blood by
accident, was able to understand the language of birds, and was told
by two of Odin’s ravens that Regin was treacherous. After slaying
Regin, Sigurd rode away with two bundles of the treasures slung
across his horse’s back. He found and awakened Brynhildr, a
beautiful woman asleep in a house on a hill. (She is known in the
different tales in which she has figured as Brynhildr, Brunhild,
Brunehault, and Brünnhilde.) The next part of the tale is most clearly
set forth in the “Nibelungenlied,” an epic poem in Middle High
German dialect, containing a story—or, more correctly, a series of
stories—which originally belonged to the entire Teutonic people.
These have been found in multitudinous poems and sagas, from
those written by the ancient Norsemen, and most primitive in form, to
the modern books, essays, and poems of writers who have been
impressed with the interesting and picturesque aspects of the
strange, complicated old story. The “Nibelungenlied” itself deals
rather with the period of Christianity—with the knights and ladies of
the time of chivalry—than with the primeval gods and heroes of the
Golden Age. The substance of its contents may be found in the
“Edda” and in the “Thidrekssaga” (thirteenth century), and the
original manuscripts of the “Nibelungenlied” itself date from the
thirteenth to the sixteenth century.
The story contained in this poem is, briefly told, as follows:
Siegfried, son of Siegmund and Sieglind, woos Kreimhild, the sister
of King Gunther, of Burgundy, promising, in return for her hand, to
aid Gunther in winning Brunhild, Queen of Issland (Iceland).
Siegfried, with the help of his cloud-cloak, conquers Brunhild for
Gunther—first in three athletic games, which she makes a test for all
suitors; and later when, after the marriage, she proves stormy and
untamed. He takes her Ring and girdle, and gives them to his wife,
Kreimhild. They possess magic properties, and Brunhild, when
deprived of them, loses her great power and becomes like any
ordinary woman. She sees her Ring on Kreimhild’s hand one day,
and, realizing that it is Siegfried, and not her husband Gunther, who
has conquered her great strength and stolen her magic circlets, she
tells her wrongs to Hagan, who promises revenge. Hagan is the
Knight of Trony, and he and his brother Dankwort are Gunther’s
vassals. Hagan entices Kreimhild to reveal to him the secret of her
husband’s safety in battle, and she tells him that Siegfried once slew
a dragon and bathed in the blood, which made him invulnerable,
save in one place, between his shoulders, where a leaf fell,
protecting the skin from the blood. Kreimhild is entirely deceived by
Hagan, and, not suspecting his treachery, she sews a circle of silk
upon her husband’s vesture over the vulnerable spot, that Hagan
may better know how to protect the hero’s one weakness when they
are in battle. It is there, where the circle of silk is sewn, that Hagan
stabs him.
There is much more in the “Nibelungenlied,” and a character famous
in poesy and sagas is introduced later in the poem—Atli, or Attila,
King of the Huns; but he has nothing to do with our story, though
some one has drawn a resemblance between his character and that
of Hunding. The “Nibelungenlied,” after Siegfried’s death, contains
very little connected in any way with Wagner’s four operas.
There are other versions of this tale, as there are of all ancient
stories. There are many tales of the killing of the Dragon and the
awakening of Brunhild, and the personality and history of the latter
have passed under diverse alterations in color and development.
One story says that Brynhildr, the Valkyrie, was made to slumber by
her father Odin, who pricked her in the temple with a sleep-thorn.
Many writers tell of a fire-circle which surrounded the sleeper and
guarded her slumbers. She is known as a great queen, a woman
gifted with magic powers, and a disobedient Walküre in different
tales; and her character changes as constantly as her history in the
various legends where we read of her. Sigurd, Siegfried, and Sinfiotli
are, in many respects, so similar that they might safely be termed
identical, though sometimes, as in William Morris’s “Sigurd, the
Volsung,” they appear as distinct characters.
Out of this confused and complicated sea of myths, legends, and old
Norse stories Wagner has drawn the material for his wonderful cycle.
His gods and goddesses are taken, with very few changes, directly
from their original place—the Teutonic mythology. His giants and
dwarfs are also unaltered as complete races. In his usage of them
he differs in some respects from the older stories.
Fafnir, the son of Rodmar, becomes the giant Fafner, and his brother
Fasolt is added. Regin is transformed into Mime, the master-smith.
Instead of Otter, who must be covered by gems, we have the love
goddess Friea, and instead of the hair which the Ring must cover in
the old legend, it is in Wagner’s adaptation one of Friea’s beautiful
eyes. Fafner hides in Hate Hole instead of upon Glistenheath, and is
killed by Siegfried instead of Sigurd. The lonely Walküres’ Rock
takes the place of the house on the hill, and instead of being made
invulnerable by the Dragon’s blood, Siegfried is protected by
Brünnhilde’s spells—a fancy which seems more poetic and beautiful,
but which originates, I believe, entirely with Wagner. Gutrune takes
the place of Kreimhild, and Hagan is not Gunther’s vassal, but his
half-brother. These are, after all, apparently slight changes, yet to
Wagner’s cycle a new poetry seems to have come. The barbaric
aspects of the tale have faded, and all the simple beauty of those
wild, noble gods and demi-gods has gleamed forth as gloriously as
the wonderful Rhinegold, which the master has made next in
importance to the gods and the dusk of their splendor.
Before going further, perhaps it might be well to say a few words of
explanation as to the motifs which form the key-notes of Wagner’s
great musical dramas.
When he set his poem of the Nibelungen Ring to music, he was not
satisfied with merely beautiful airs and harmonies linked together
with no purpose save the lovely sounds. He wished, above all, to
have his music fit his words; and for every character and thought and
incident, and indeed for almost everything in his operas, he wrote a
melody, and these descriptive musical phrases are called motifs.
Each one has its meaning, and when it is played it brings the thought
of what it describes and represents, and it makes a double language
—what the characters on the stage are saying and what the music is
saying, as well. Through the motifs we understand many things
which we could not possibly comprehend otherwise.
That Wagner wished to give the impression that Erda was the
mother of all beings, divine and human, at the beginning of the
world, he has shown by the fact that the motif of the Primal Element
—the commencement of all things—is identical with hers, save that
where she is indicated the melody takes a minor coloring, denoting
her character of mystery as well as the gloom in which her prophetic
powers must necessarily envelop her. The contrasting, yet
harmonizing, elements of earth and water are also shadowed forth, I
think, in this motif of the Primal Element, which is used for the Rhine,
and also for the Goddess of the Earth. When the Vala’s daughters—
the Nornir—are mirrored in the music, the same melody appears,
fraught with the waving, weaving sound of their mystic spinning.
The motifs in Wagner’s operas are, above all, descriptive. For
example, note the Walhalla, Nibelung, and Giant motifs.
The first of these, full of power, substance, and dignity, not only is
descriptive of the great palace itself, but also represents the entire
race of gods who inhabit it, seemingly secure in their conscious glory
and sovereignty. To indicate Wotan, the King of the gods and the
ruler in Walhalla, Wagner has constantly made use of this motif.
Its melody is measured, strong, and simple, and the nobility of those
worshipped gods of primeval years seems to breathe through it.
The Nibelungs were so intimately associated with their work that
they were scarcely more than living machines—soulless exponents
of the art of the forge and the anvil; so when we hear in the music
the beat of hammers—the sharp, metallic clang in measured time,
our first thought is that the hammers are swung by the Nibelungs.
How cramped is their melody, how monotonous and hopeless is the
regular fall of the hammers! When we hear it hushed and veiled with
discords, we seem to come in contact with the narrow, darkened
souls of the Nibelungs.
And now we come to the motif of the giants.
It is, like themselves, heavy, lumbering, with a slur that is like the
stumbling of heavy feet. Clumsy and ungraceful, it and what it
represents cross the idyllic beauty of the motifs of Friea, Walhalla,
the Ring, the Rhinegold, and the rest, with a harsh and disagreeable
sense of an inharmonious element. How different from the majestic
gods, and the clever, small-souled Nibelungs, are these great
creatures who are all bodies and no brains, and who are so ably
represented by the music allotted them in the operas! Yet, in their
own way, they and their motif are extremely picturesque!
In these three motifs we can see the genius which formed them, and
so many others, even greater in conception and execution. Scattered
throughout The Story of the Rhinegold will be found a few of these
motifs—only a few and not the most lovely—but enough I think to
help one, in a small way, to follow the operas with more interest and
understanding than if one did not know them.
One of the simplest motifs in the book is one of the most important:
the Rhinegold motif. It is like the blowing of a fairy horn heralding to
the world of sprites and elves the magic wonder in the river.
In the olden days they had a lovely legend of the formation of the
Rhinegold. They said that the sun’s rays poured down into the Rhine
so brilliantly every day that, through some magic—no one knew
exactly how—the glowing reflection became bright and beautiful
gold, filled with great mystic powers because of its glorious origin—
the sunshine.
And that was the beginning of the Rhinegold.
A.A.C.
 CONTENTS

Part I
THE RHINEGOLD, or DAS RHEINGOLD
page
Prelude 3
chapter
I. The Rhine Maidens 8
II. Fasolt and Fafner 13
III. Nibelheim 18
IV. The Rainbow Bridge 24

Part II
THE WARRIOR GODDESS, or DIE WALKÜRE
Prelude 33
chapter
I. The House of Hunding 37
II. The Daughter of Wotan 45
III. Brünnhilde’s Punishment 54

Part III
SIEGFRIED
Prelude 63
chapter
 I. Siegfried and Mime 67
II. Hate Hole 79
III. The Mountain Pass 88
IV. The Walküres’ Rock 95

Part IV
THE DUSK OF THE GODS, or GÖTTERDÄMMERUNG
Prelude 103
chapter
I. The Hall of the Gibichungs 107
II. The Walküres’ Rock Once More 113
III. The Rhine Chief’s Bride 118
IV. On the Banks of the Rhine 124
V. The Last Twilight 133
 ILLUSTRATIONS

WOTAN AND BRÜNNHILDE Frontispiece

Facing p.
THE GLEAMING TREASURE
10
A WARRIOR GODDESS 34
THE WALKÜRE APPEARS 50
SIEGFRIED AT THE FORGE 76
THE DEATH OF THE DRAGON 82
BRÜNNHILDE ON THE WALKÜRES’
104
ROCK
GUTRUNE AND SIEGFRIED 110
BRÜNNHILDE AND SIEGFRIED 116
GUNTHER AND BRÜNNHILDE 122
HAGEN AND SIEGFRIED 128
AFTER SIEGFRIED’S DEATH 130
 Part I

THE RHINEGOLD, or DAS RHEINGOLD

Motif of the Rhinegold

[Listen] [MusicXML]

PRELUDE

We have, all of us, read of the Golden Age, when the gods ruled
over the world, and giants and dragons, dwarfs and water-fairies
inhabited the earth and mingled with mortals. The giants were then a
strong, stupid race, more rough than cruel, and, as a rule, generous
among themselves. They were very foolish creatures, and constantly
did themselves and others harm; but their race, even at that time,
was dying out, and there were left of it only two brothers, Fasolt and
Fafner.
The dwarfs, or Nibelungs, were entirely different. They were small
and misshapen, but very shrewd, and so skilful were their fingers
that they were able to do the most difficult work in the finest metals.
They lived in an underground country called Nibelheim (Home of the
Welcome to our website – the ideal destination for book lovers and
knowledge seekers. With a mission to inspire endlessly, we offer a
vast collection of books, ranging from classic literary works to
specialized publications, self-development books, and children's
literature. Each book is a new journey of discovery, expanding
knowledge and enriching the soul of the reade

Our website is not just a platform for buying books, but a bridge
connecting readers to the timeless values of culture and wisdom. With
an elegant, user-friendly interface and an intelligent search system,
we are committed to providing a quick and convenient shopping
experience. Additionally, our special promotions and home delivery
services ensure that you save time and fully enjoy the joy of reading.

Let us accompany you on the journey of exploring knowledge and

personal growth!

ebookultra.com