International Journal of Trend in Scientific Research and Development (IJTSRD)

Volume 5 Issue 2, January-February 2021 Available Online: www.ijtsrd.com e-ISSN: 2456 – 6470

Traditional Machine Learning and No-Code

Machine Learning with its Features and Application
Hiteshkumar Babubhai Vora, Hardik Anilbhai Mirani, Vraj Bhatt
Information Technology Nirma University, Ahmedabad, Gujarat, India

ABSTRACT How to cite this paper: Hiteshkumar

This is the new era of technology development where all the things and work Babubhai Vora | Hardik Anilbhai Mirani |
is done by the machines. The goal of Information Technology is to develop a Vraj Bhatt "Traditional Machine Learning
device which is able to work like a human itself. For that Artificial Intelligence, and No-Code Machine Learning with its
Machine Learning and Deep Learning are going to be used. Machine Learning Features and Application" Published in
is a subpart of the Artificial Intelligent which helps a machine to learn by itself. International Journal
To apply learning processes on machines it required deep knowledge of of Trend in Scientific
programming, mathematics and statistics. Now it is not a big problem, as the Research and
technology is changing day by day the new concept known as No-Code ML and Development (ijtsrd),
Auto-Code Generation are introduced. This helps the users to create a model ISSN: 2456-6470,
without doing any kind of coding. In this new technology everyone is able to Volume-5 | Issue-2,
create a model and use machine learning. There are several platforms which February 2021, IJTSRD38287
provide this kind of facilities. The models created on those platforms give good pp.29-32, URL:
accuracy and desire outcomes as well. www.ijtsrd.com/papers/ijtsrd38287.pdf

KEYWORDS: Auto-Code Generation, Deep Learning, Artificial Intelligent, Auto Copyright © 2021 by author(s) and
algorithm selection, No-Code ML platforms International Journal of Trend in Scientific
Research and Development Journal. This
is an Open Access article distributed
under the terms of
the Creative
Commons Attribution
License (CC BY 4.0)
(http://creativecommons.org/licenses/by/4.0)

INTRODUCTION
With this fast developing world computers play a vital role in needs to train a model by applying the most appropriate
almost every sector. Machine can work as an instruction algorithm on the kind of data. In today’s world the data
given to it. Without giving any instruction to it, it becomes generated is not in structure or the data which is the most
useless. But with the concept of Artificial Intelligent now appropriate in to train the model so that before applying any
machines are becoming smart enough that they can work on algorithm on the data it is required to arrange the data in the
their own. This can be achieved through the concepts of most appropriate manners.
machine learning. Taking an example of allowing only
authorized persons in the organization, if unauthorized As shown in figure 1, ML follows the above flow. First task is
persons try to enter then generate and alert. But as human to import the data on which the model is needed to create.
faces change slowly day by day it is not possible to recognize The data can be in any form i.e. images, csv, text, excel etc.
the person without help of machine learning. In this situation The second task is to create structured data. We have to give
machine learning helps to do the tasks. all the data in some specific structure so that the model can
be trained more accurately. After having the structure data,
In this era machine learning has many impacts on the it might be possible that all data are not present or some of
different real time areas. With the development of new the information is missing in the data. It also has some
technology and languages like Python, machine learning is outliers which affect the entire model. So it is necessary to
becoming easy yet it requires more and deep knowledge of remove such data points. This is done using the
statistics and programming as well. With introduction to No- preprocessing of the data. In this stage feature engineering,
Code ML, it becomes easy to create a model without having feature transformation, feature selection, missing values
knowledge of the programming and statistics. filling, outlier handling, checking for variable types etc is
done on this stage. In other phases the model/algorithm is
TRADITIONAL MACHINE LEARNING APPROACH selected from the library or the new algorithm is applied on
Machine learning is a way to train a machine to predict the the data. By checking the type of the data proper model is
future results. In traditional machine learning approach it selected. Then model apply on the data and train the model.

@ IJTSRD | Unique Paper ID – IJTSRD38287 | Volume – 5 | Issue – 2 | January-February 2021 Page 29

 International Journal of Trend in Scientific Research and Development (IJTSRD) @ www.ijtsrd.com eISSN: 2456-6470
and ask the question in the English. The model will
automatically train. Then the user can get his output and
evaluate the model.

Figure 1: Approach of traditional machine learning

Before training the model partition is done i.e. the train data
and validates data. After training and validation the data
metric and score model is created. If the score is too less or
not reached to the targeted score then the preprocessing,
model applying, model assessment stages are repeated.
When the model reaches the targeted or enough accuracy
then at the next stage the tested data is also added in the
data set and the model is re-trained with the new data which
is predicted by model. Then in the next stage the model is
deployed and the maintenance of the model is done by
taking all the information of the model's output. If any
further need is created then all the working and predicted
data and results are added into the data set and all the Figure 2: Approach of No-Code ML with traditional
process from training the model is repeated. On the above approach
way in every stage a good knowledge of programming and
statistics is needed. But the problem with this is that all WORKING OF ALGORITHM
statistics specialists don’t know programming and all the It looks very simple that the model is created automatically
programmers are not experts in the field of statistics. So as but in the background many processes are done by the
the solution No-Code ML is introduced. platforms. Those stages are described here.

NO-CODE ML APPROACH 1. Preprocessing

No-Code ML is a process of creating a model with minimum After selecting the database we press the GO button and then
knowledge of programming and statistics skills. It is an work on the algorithm is started. In the first phase it
Artificial Intelligent base algorithm which works for the analyzes the dataset. It removes all the rows which have the
coding and accuracy calculation. The No-Code ML works as NULL value or in some cases it fills the NULL value with the
described in figure 2. mean value of that column. It also analyzes all the non
numerical columns and removes the column which is not
As shown in figure it becomes too simple to create a model in required for the model. It also finds for the outliers in the
No-Code ML platform. It is an Artificial Intelligent base data and removes all the outliers if any. In some kind of
algorithm which selects the best ML algorithm to apply on dataset it requires to perform the normalization as well. By
the model. It is just a drag and drop process. In any No-Code doing it at the first stage, it increases the chances of creating
ML platform, a user needs to drag and drop his/her dataset a more accurate model.

@ IJTSRD | Unique Paper ID – IJTSRD38287 | Volume – 5 | Issue – 2 | January-February 2021 Page 30

 International Journal of Trend in Scientific Research and Development (IJTSRD) @ www.ijtsrd.com eISSN: 2456-6470
2. Training Model There are several other platforms which facilitate the same
In this phase many combinations and permutations are features.
applied on the dataset and algorithm. After applying the
many algorithms as per the data set accuracy is checked for USE CASES & PERSPECTIVE
the many thousands of algorithms. At the end of this phase it Organizations having Less Potential Programmers or Having
gives a model with the most accurate model. Then it runs it No team of Data Scientist or ML Engineers, can Guide the
as a final algorithm on a given data set and stores the model. Sales and marketing team / Analyst team / R & D team and
All this process takes some time to find the most accurate they will be able to put their efforts for such things like
model for the dataset. predictions, classification Etc. Using No- Code ML approach
the Person will be able to solve complex Data Science and ML
3. Testing for accuracy problems without knowledge of Python or R programming.
After creating the model AI test the model and find the final The main purpose and Use Case of No-code ML is to fill the
accuracy of the model. This testing is done on the partial gap between skilled Programmers and skilled statisticians.
dataset as well. For example if there is one thousand data
rows are there in the dataset then the model accuracy found Area of Use:
on all the batches of hundreds rows for the whole dataset. It Various industries are Using and Approaching No-code ML,
ensures the accuracy for all data in the dataset. Real estate, Finance, Retail & consumer goods, Technology,
Insurance, Healthcare, Fintech, Transportation, banking Etc.
At this point of time No-Code ML works only for regression
and classification algorithms only but in the near future it STATISTICS OF SAMPLE PROGRAM
will adopt all other algorithms of Machine Learning. Dataset
Class Image Dataset
TOOLS OF NO-CODE ML Dog 252
There are many tools available for performing the No-Code Cat 252
ML. In the new era of computer science and technology many Table: 1 Information of Dataset
businesses want software based on AI. For that they need to
hire the software engineer and AI experts which cost the Class Accuracy
organization a lot, so the solution of that No-Code ML Samples are taken 38 per each class
platform was introduced. Some of them are described here.
Class Accuracy
1. Create ML Dog 0.89
Create ML is Apple's no-code drag and drop tool. It is an Cat 0.95
independent macOS application which comes with a bunch Table: 2 Accuracy achieve by the model
of pre-trained models. It allows us to use transfer learning to
build our own models. It covered all the algorithms for
Natural Language Processing, Recommendation System etc…
It provides real-time results on validation data for a model.
The model can be deployed on any iOS application.

2. Google AutoML
This AutoML tool is developed by the world’s leading
company GOOGLE. Google provides the same feature as
CreateML. AutoML currently includes Vision i.e. image
classification, Natural Language Processing, Video Intelligent
etc… This tool enables the users to train models on specific
to use cases. It allows exporting the model in .pb or .tflite
formation. Figure 3: Matrix of the model accuracy

3. Teachable Machine CONCLUSION

It is also a platform provided by Google. It allows training There are so many factors where there is a need to complete
models quickly to recognize images, sounds, and poses right the work which can be done more easily and more quickly by
from the browser. It allows directly dragging and dropping using the machine learning algorithm like Financial and
files to teach the model or use a web camera to do this. This Retail Industries. But every person or employee cannot
platform uses the library Tensorflow.js. This platform allows implement a Machine learning algorithm without knowing
exporting files in Tensorflow.js or tflite formats. about programming language. But as human nature, we
always try to find a way to decrease complexity. And so that
4. Microsoft Azure Automated Machine Learning giant technical companies have developed many platforms
Microsoft Azure is a tool which is capable of developing where anyone can use machine learning without knowing a
models at a faster pace. This tool automatically deploys programming language. as we know that every coin has two
predicted models with some existing data which are already sides. So the NoCodeML concept also has a negative side that
filtered with some algorithms. This tool is also able to detect it becomes hard to give accurate results when the approach
error in data and rectifies the data. It also gives metrics of machine learning is unsupervised data and Reinforcement
visualization enabling a user to make comparisons between Learning. But this topic is in their Initial stage. We can use
various models. this approach in every less-complex machine learning work.

@ IJTSRD | Unique Paper ID – IJTSRD38287 | Volume – 5 | Issue – 2 | January-February 2021 Page 31

 International Journal of Trend in Scientific Research and Development (IJTSRD) @ www.ijtsrd.com eISSN: 2456-6470
REFERENCES [4] Z. Dai, Z. Yang, Y. Yang, W. W. Cohen, J. Carbonell, Q. V.
[1] K. He, X. Zhang, S. Ren, J. Sun, Deep residual learning Le, R. Salakhutdinov, Transformer-xl: Attentive
for image recognition, in: Proceedings of the IEEE language models beyond a fixed-length context, arXiv
conference on computer vision and pattern preprint arXiv:1901.02860
recognition, 2016, pp. 770–778.
[5] Teachable machines for accessibility ; November
[2] Xin He, Kaiyong Zhao, Xiaowen Chu, AutoML: A 2017 ; ACM SIGACCESS Accessibility and Computing
Survey of the State-of-the-Art, arXiv:1908.00709v5
[6] https://www.kdnuggets.com/2019/03/why-automl-
[cs.LG] 8 Jul 2020
wont-replace-data-
[3] C. Gong, D. He, X. Tan, T. Qin, L. Wang, T.-Y. Liu, Frage: scientists.html#:~:text=As%20a%20result%2C%20it
frequency-agnostic word representation, in: Advances %20is,do%20not%20cover%20this%20approach.
in Neural Information Processing Systems, 2018, pp.
1334–1345.

@ IJTSRD | Unique Paper ID – IJTSRD38287 | Volume – 5 | Issue – 2 | January-February 2021 Page 32