Principles and Theories

of Data Mining With

RapidMiner

Sarawut Ramjan
Thammasat University, Thailand

Jirapon Sunkpho
Thammasat University, Thailand

A volume in the Advances in Computer and

Electrical Engineering (ACEE) Book Series
Published in the United States of America by
IGI Global
Engineering Science Reference (an imprint of IGI Global)
701 E. Chocolate Avenue
Hershey PA, USA 17033
Tel: 717-533-8845
Fax: 717-533-8661
E-mail: [email protected]
Web site: http://www.igi-global.com

Copyright © 2023 by IGI Global. All rights reserved. No part of this publication may be reproduced, stored or distributed in
any form or by any means, electronic or mechanical, including photocopying, without written permission from the publisher.
Product or company names used in this set are for identification purposes only. Inclusion of the names of the products or
companies does not indicate a claim of ownership by IGI Global of the trademark or registered trademark.
Library of Congress Cataloging-in-Publication Data
Names: Ramjan, Sarawut, 1984- author. | Sunkpho, Jirapon, DATE- author.
Title: Principles and theories of data mining with RapidMiner / authored by
Sarawut Ramjan, and Jirapon Sunkpho.
Description: Hershey, PA : Engineering Science Reference, [2023] | Includes
bibliographical references and index. | Summary: “This book is
academically written as a guide for students and people interested in
experimenting Data Mining using RapidMiner software. It covers the
contents related to Data Mining, which consists of Classification, Deep
Learning, Association Rule, Clustering, Recommendation System and
RapidMiner Software usage as well as researching case studies on the use
of data mining techniques in data science. Additionally, this book is
the foundation of Python programming for data science for young
scientists who want to understand data mining algorithms. As well as
starting to write programs that can be applied to other data science
programs. At the end of this book, authors describe about data
governance with a case study of the government sector to enable young
data scientists to understand the role of data scientists as part of
stakeholders in data governance actions. The authors hope that this book
is a good beginning for those who would like to develop themselves or
for those who own data within their organization to meet internal and
external problems. RapidMiner software is used to analyze data and
provide guidance for further study in data science at a higher level”--
Provided by publisher.
Identifiers: LCCN 2022061153 (print) | LCCN 2022061154 (ebook) | ISBN
9781668447307 (h/c) | ISBN 9781668447314 (s/c) | ISBN 9781668447321
(eISBN)
Subjects: LCSH: Data mining. | Big data. | RapidMiner (Electronic
resource).
Classification: LCC QA76.9.D343 R356 2023 (print) | LCC QA76.9.D343
(ebook) | DDC 006.3/12--dc23/eng/20230123
LC record available at https://lccn.loc.gov/2022061153
LC ebook record available at https://lccn.loc.gov/2022061154

This book is published in the IGI Global book series Advances in Computer and Electrical Engineering (ACEE) (ISSN:
2327-039X; eISSN: 2327-0403)

British Cataloguing in Publication Data

A Cataloguing in Publication record for this book is available from the British Library.

All work contributed to this book is new, previously-unpublished material. The views expressed in this book are those of the
authors, but not necessarily of the publisher.

For electronic access to this publication, please contact: [email protected].

 Advances in Computer and
Electrical Engineering (ACEE)
Book Series
Srikanta Patnaik
SOA University, India
ISSN:2327-039X
EISSN:2327-0403
Mission
The fields of computer engineering and electrical engineering encompass a broad range of interdisci-
plinary topics allowing for expansive research developments across multiple fields. Research in these
areas continues to develop and become increasingly important as computer and electrical systems have
become an integral part of everyday life.
The Advances in Computer and Electrical Engineering (ACEE) Book Series aims to publish
research on diverse topics pertaining to computer engineering and electrical engineering. ACEE encour-
ages scholarly discourse on the latest applications, tools, and methodologies being implemented in the
field for the design and development of computer and electrical systems.

Coverage
• Computer Hardware
IGI Global is currently accepting manuscripts
• Microprocessor Design
for publication within this series. To submit a pro-
• Electrical Power Conversion
posal for a volume in this series, please contact our
• Power Electronics
Acquisition Editors at [email protected]
• Algorithms
or visit: https://www.igi-global.com/publish/.
• Applied Electromagnetics
• Computer Architecture
• Digital Electronics
• Programming
• Optical Electronics

The Advances in Computer and Electrical Engineering (ACEE) Book Series (ISSN 2327-039X) is published by IGI Global, 701 E.
Chocolate Avenue, Hershey, PA 17033-1240, USA, www.igi-global.com. This series is composed of titles available for purchase individually;
each title is edited to be contextually exclusive from any other title within the series. For pricing and ordering information please visit https://
www.igi-global.com/book-series/advances-computer-electrical-engineering/73675. Postmaster: Send all address changes to above address.
Copyright © 2023 IGI Global. All rights, including translation in other languages reserved by the publisher. No part of this series may be
reproduced or used in any form or by any means – graphics, electronic, or mechanical, including photocopying, recording, taping, or informa-
tion and retrieval systems – without written permission from the publisher, except for non commercial, educational use, including classroom
teaching purposes. The views expressed in this series are those of the authors, but not necessarily of IGI Global.
 Titles in this Series
For a list of additional titles in this series, please visit: www.igi-global.com/book-series

NeutroGeometry, NeutroAlgebra, and SuperHyperAlgebra in Today’s World

Florentin Smarandache (University of New Mexico, USA) and Madeline Al Tahan (Abu Dhabi University UAE)
Engineering Science Reference • © 2023 • 300pp • H/C (ISBN: 9781668447406) • US $270.00

Energy Systems Design for Low-Power Computing

Rathishchandra Ramachandra Gatti (Sahyadri College of Engineering and Management, India) Chandra Singh
(Sahyadri College of Engineering and Management, India) Srividya P. (RV College of Engineering, India) and
Sandeep Bhat (Sahyadri College of Engineering and Management, ndia)
Engineering Science Reference • © 2023 • 389pp • H/C (ISBN: 9781668449745) • US $270.00

Artificial Intelligence Applications in Battery Management Systems and Routing Problems in Electric Vehicles
S. Angalaeswari (Vellore Institute of Technology, India) T. Deepa (Vellore Institute of Technology, India) and L.
Ashok Kumar (PSG College of Technology, India)
Engineering Science Reference • © 2023 • 342pp • H/C (ISBN: 9781668466315) • US $260.00

Handbook of Research on AI Methods and Applications in Computer Engineering

Sanaa Kaddoura (Zayed University, UAE)
Engineering Science Reference • © 2023 • 632pp • H/C (ISBN: 9781668469378) • US $335.00

Structural and Functional Aspects of Biocomputing Systems for Data Processing

U. Vignesh (Information Technology Department, Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science
and Technology, Chennai, India) R. Parvathi (Vellore Institute of Technology, India) and Ricardo Goncalves
(Department of Electrical and Computer Engineering (DEEC), NOVA School of Science and Technology, NOVA
University Lisbon, Portugal)
Engineering Science Reference • © 2023 • 246pp • H/C (ISBN: 9781668465233) • US $270.00

5G Internet of Things and Changing Standards for Computing and Electronic Systems
Augustine O. Nwajana (University of Greenwich, UK)
Engineering Science Reference • © 2022 • 308pp • H/C (ISBN: 9781668438558) • US $250.00

Theory and Applications of NeutroAlgebras as Generalizations of Classical Algebras

Florentin Smarandache (University of New Mexico, USA) and Madeline Al-Tahan (Lebanese International Uni-
versity, Lebanon)
Engineering Science Reference • © 2022 • 333pp • H/C (ISBN: 9781668434956) • US $270.00

701 East Chocolate Avenue, Hershey, PA 17033, USA

Tel: 717-533-8845 x100 • Fax: 717-533-8661
E-Mail: [email protected] • www.igi-global.com
Table of Contents

Preface................................................................................................................................................... vii

Chapter 1
Introduction to Data Mining ...................................................................................................................1

Chapter 2
Data .......................................................................................................................................................35

Chapter 3
Software Installation and Introduction to RapidMiner .........................................................................50

Chapter 4
Data Pre-Processing and Example of Data Classification With RapidMiner .......................................65

Chapter 5
Classification .........................................................................................................................................83

Chapter 6
Deep Learning.....................................................................................................................................107

Chapter 7
Clustering ............................................................................................................................................139

Chapter 8
Association Rule .................................................................................................................................160

Chapter 9
Recommendation System ....................................................................................................................177

Chapter 10
Case Studies on the Use of Data Mining Techniques in Data Science ...............................................193


 

Chapter 11
Data Mining for Junior Data Scientists: Basic Python Programming ................................................205

Chapter 12
Data Mining for Junior Data Scientists: Data Analytics With Python ...............................................237

Glossary .............................................................................................................................................293

Compilation of References ...............................................................................................................304

About the Authors .............................................................................................................................317

Index ...................................................................................................................................................318
 vii

Preface

Welcome to this guide on data mining using RapidMiner software. This comprehensive book is specifi-
cally designed for beginners and anyone who wants to explore the world of data mining. The book is
structured into 12 chapters, each covering a specific topic in detail.
Chapter 1, “Introduction to Data Mining,” gives an insight into the trend and importance of data
mining in various industries. Chapter 2, “Data,” explains the different types of data that a data scientist
should understand before analyzing it using data mining techniques.
Chapter 3, “Software Installation and Introduction to RapidMiner,” provides a detailed guide for
beginners who want to practice using RapidMiner. Chapter 4, “Data Pre-Processing and Example of
Data Classification With RapidMiner,” guides a beginner in understanding data pre-processing and how
to apply it using RapidMiner.
Chapter 5, “Classification,” enables beginners to apply RapidMiner to classify data that answers
business questions. Chapter 6, “Deep Learning,” introduces the concept of deep learning and provides
practice exercises using RapidMiner to apply deep learning techniques for data analytics.
Chapter 7, “Clustering,” teaches beginners how to apply RapidMiner to divide data by clustering
techniques. Chapter 8, “Association Rule Mining,” introduces the creation of association rules from a
dataset by RapidMiner.
Chapter 9, “Recommendation System,” explains the concept of recommendation systems and how to
apply them with RapidMiner. The chapter 10, titled “Case Studies on the Use of Data Mining Techniques
in Data Science,” furnishes the novice with the author’s scrutinized data mining technique outcomes,
thus facilitating the application of data mining to research domains.
Chapters 11 and 12, “Python Programming for Data Mining,” introduce beginners who want to
analyze data with more complex data mining techniques.
This book is ideal for beginners who want to grow in their data scientist career or anyone who wants
to address internal and external issues using data in their organization. Get ready to dive into the exciting
world of data mining using RapidMiner software with this guide.

Sarawut Ramjan
Thammasat University, Thailand

Jirapon Sunkpho
Thammasat University, Thailand


 1

Chapter 1
Introduction to Data Mining

ABSTRACT
Data mining is a powerful and increasingly popular tool that uses machine learning to uncover pat-
terns in data and help businesses stay competitive. Data scientists are trained to understand business
objectives and select the correct techniques for data exploration and pre-processing. After formulating
the business question, data mining methods are chosen and evaluated to determine their ability to fit the
data set and answer the query. Results are then reported back to the business owner. Data mining is an
essential part of modern business, allowing the organization to keep up with the competition and remain
successful. With its growing popularity, the need for data scientists is rapidly increasing.

INTRODUCTION

Nowadays, data analysis using data mining is becoming more and more important. The digital ecosystem
supports the storage of massive amounts of data either Cloud Technology or Database with large amount
of memory. Moreover, these data are generated with users around the world in real time with high the
speed communication through Social Networks resulting in a massive growth of data, known as Big Data.
Businesspeople around the world can take advantage of it. For instance, to analyze customer purchasing
patterns or predict the number of raw materials for production, it requires Data Mining techniques to
analyze the patterns and predict the outcomes necessary for entrepreneurs’ decision making.
Data analysis for business decision making can be done with various techniques such as Descrip-
tive Statistics and Inferential Statistics. Therefore, the Data Mining technique is not a replacement for
traditional data analysis, but it should be considered the development of advanced analysis techniques
along with the work of machine learning technology. It supports the processing of large amount of data,
Big Data. Consequently, Data Mining has yielded accurate results consistent with business questions.

DOI: 10.4018/978-1-6684-4730-7.ch001

Copyright © 2023, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
 
Introduction to Data Mining

IMPLEMENTATION OF DATA MINING AND CHALLENGES

With the need for problem analysis, it has created a position to be responsible for data analysis, namely
Data Scientist. However, the Data Scientists are not solely from experts in digital technology, but they
can be any person who owns the data. The medical technicians, who have information on the patients’
health and understand the context of medical industry, question asking and the nature of the data, can
become a Data Scientist.
Many industries have employed data scientists who are not familiar with the industry or lack expe-
riences in the data they are to analyze. Therefore, the data scientists have to consult the questioners,
possibly from the management department, and the experts in order to obtain results consistent with the
facts. Therefore, in the scientific work, the data can be applied in various industries as presented in the
following examples.

• In the market, to determine the selling price, the consumers can see a model which consists of
many sub-models. The prices vary depending on the car’s components. However, the consumers
can still find similar pricings in all sub-models. For instance, Sub-model 1 costs 1,000,000 baht.
The 2nd Sub-model is priced at 1,600,000 baht and the 3rd Sub-model is priced 1,7000,000 baht.
It is obvious that the difference between the 2nd and 3rd Sub-models is only 100,000 baht. Such
a price setting is caused by dividing customers into 3 groups to suit the number of Sub-models
which is three. Data mining techniques are then used to analyze the mid-prices of each customer
segment of each Sub-model. Unless the data scientists analyze the data using the techniques, gen-
eral businesspeople may divide the pricing into 1,000,000 baht, 1,300,000 and 1,700,000 baht for
Sub-data respectively. The clustering techniques are, therefore, used to determine the price of car
sales to be able to set a suitable price for each target group. And the dealers also get the most profit.
• To forecast the condominium price in Bangkok (Sunkpho and Ramjan, 2020) using data analysis,
it was found that variables affecting the prices of the condominium in Bangkok are the distance
from the condo to Skytrain and MRT stations, the number of rooms, the number of floors and the
age of the condominiums. Data scientists use deep learning techniques to analyze such variables
and the prices, and found that the smaller distance from the condo to the sky train and subway sta-
tions, the higher the prices are. The more numbers of rooms are, the lower the prices will become.
The more floors the condos have, the higher the prices are. And if the age of the condominium is
less, its price is high. Therefore, the condominium real estate industries can consider such vari-
ables to determine the appropriate prices of the condominium projects.
• To categorize borrowers with the ability to pay debts, Banks have to classify their borrowers. To
promote bank services, the borrowers are convinced to extend their loans with attractive offers
such as a lower interest rate (Refinance). Data scientists can use data mining techniques to classify
the customers by analyzing various variables. Since the banks have information about borrowers
such as age, income, loan duration, default rate and the amount of additional loans that have al-
ready been approved, they can classify borrowers with the ability to repay their debts, so that they
can offer marketing promotions. Banks can increase their interest income further.

From the sample cases, Data scientists do not only need to develop their knowledge on data analysis
with data mining techniques, but have to study on statistics, database technology, data visualization and

2

Introduction to Data Mining

knowledge relevant to the data analytics industry. Although data science offers benefits to a variety of
industries, these industries still face dimensional the following challenges (Tanantong, 2020).
Scalability: The size of the data keeps growing from the updates of social network users or from the
updates of business partners (Almasoud, Al-Khalifa and Al-Salman, (2019). For example, travel agents
from around the world update their booking information through the Global Distribution System (GDS)
at the same time and make transactions round the time. As a result, the databases grow larger by users
all the time. Therefore, in the perspective of the industries that need to use big data for analysis, the
algorithms in data management, accessing information, transferring data over a high-speed network as
well as data mining techniques can support large-scale data processing.
High Dimensionality: With information technology, whether in the forms of photos or videos file
with a higher resolution, it can detect the nature of obstacles. When combined, these autonomous vehicle
pictures can generate huge amounts of high-resolution data (Lawal, Ibrahim, Sani and Yaakob, 2020).
Consequently, the data scientists are required to improve their algorithms in Data Exploration and Pre-
Processing for more accurate analysis.
Various and Complex Data: Since organizations store information in a data warehouse, the ware-
house contains a large number of Attributes and Data Types because the data created by each social
network platform varies depending on its design (Berman, 2018). Therefore, it is a great challenge for
data scientists to collect those data and select only the attributes useful for the analysis. Moreover, to
prepare data from a variety of sources, the data scientists need to optimize data and shape them into the
same standards appropriate for each type of data mining technique.
Data Ownership and Distribution: At present, organizations around the world store data through
cloud technology with the distributed storage system. As a result, the source of data is spread all over the
world as well. Accessing large volumes of data requires high-speed and highly secured public networks
(Jiang, Jiang and Wang, 2020). Therefore, the security of data retention and each country’s data process-
ing rights have become a challenge for data scientists. In Thailand, where the Personal Data Protection
Act is enforced, data scientists cannot take customer data for analysis without consent.

DATA ANALYSIS PROCESS WITH DATA MINING TECHNIQUES

Data Mining Technique is to analyze the data and search for data patterns from large datasets (Bramer,
2016). It is divided into four main categories; Classification, Numerical Prediction, Association Rule, and
Clustering. These analytical techniques are not recently emerging but have been used for data analysis
in order to obtain results in various forms of decision making. The main difference between data mining
techniques and the descriptive and inferential statistical analysis is the use of Machine Learning (ML)
techniques (Bell, 2015: Chiarini, Kohli and Forsgren, 2021). As a result, data mining techniques are
able to analyze data until the results are closer to the facts. In addition, the development has resulted in
a new forms of data processing such as Deep Learning which is suitable for analyzing large volumes of
data and non-linear or Association Rule Mining. In turn, data scientists can apply the model to analyze
data in accordance with business problems. Therefore, data mining techniques are used in various forms
of knowledge discovery. For example, it can be used to analyze tumor patterns that can progress to be-
come cancer. It can analyze the customers who have the ability to pay off loans, suitable for marketing
promotions and expansion to new credit lines.

3
 
Introduction to Data Mining

Data analysis process with data mining techniques in the form of Knowledge Discovery Process
(Cardona, Gomez and Trujillo, 2014: Kyaw and Limsiroratana, 2019: Gupta, Sahayadhas and Gupta,
2020) consists of:

• Data Sources: Data scientists can access multiple data sources at the same time to determine
the likelihood that those datasets can be used to analyze in order to obtain the results suitable for
answering the questions.
• Data Store: Data scientists may collect data sets from a single source, in-house database, multiple
sources, business partners and social networks such as Twitter. Data storage can be in a single data
or a data set format, formed by merging datasets from multiple data sources.
• Data Selection and Pre-Processing: Data scientists can use data processing techniques to select
data. For example, the relationship between variables in the data set and the target variables is
analyzed by seeking expert advice on the data set. Then the data are prepared for processing with
data mining techniques such as missing values, eliminating outliers.
• Data Mining: Data scientists need to choose the appropriate data mining techniques in order
to obtain results that can be used to answer the questions. In the step, the data scientist can also
review whether the data mining techniques are suitable for analyzing the existing datasets and
whether it can answer the questions. Unless the dataset and data mining techniques are consistent,
the data scientists can consider two scenarios; modifying the new data analysis method to replace
the old method but obtaining the results that can be used interchangeably or considering changing
the data set suitable for data analysis
• Interpretation: The results obtained from the analysis with data mining techniques still seem to
be quantitative data or Data Visualization (Samasiri, 2020). Data scientists need to interpret the
results so that questioners can understand it. Therefore, when asking those who wish to use the
data, Data scientists need to firstly determine whether those questions require data mining tech-
niques, or simply using the raw data to process is enough to answer those questions.

However, the Knowledge Discovery Process is generally used for data analysis. Cross-industry Stan-
dard Process for Data Mining: CRISP-DM was developed for the purpose of analyzing business-related
data (SPSS Inc, 2000).

• Business Understanding: Initially, data scientists work with business questioners; a senior man-
agement or a domain expert; to define problems in which data visualization or data mining tech-
niques can be used for answering a set of questions related to the business context. In the medi-
cal or sports industries all of which have different contexts, data scientists need to consider the
internal environment within the organization such as budgets, human resources that support data
analysis and the amount of time to analyze the data. The success of the data analysis should be
set so that when it is completed, it can indicate whether the obtained data can be used to answer
business questions. Finally, the technology used for both hardware and software analysis can be
selected. For example, if the given question requires large data processing, data scientists need to
use high-performance hardware computing to shorten the processing time. If they would like to
reduce the time to create a model for data analysis, the ready-made software can be applied. But it
costs higher compared to the data analysis by programming in open-source languages.

4

Introduction to Data Mining

• Data Understanding: It is to verify whether existing data sets, either within the business or out-
side the business, such as customer feedback on social networks, can be used to analyze the data.
In this process, the data scientists and the questioners can work together to ensure that the data
therein is accurate and consistent with the business context. In turn, it can be used for analysis
to answer business questions accurately and consistently. At this stage, the data scientists must
consult with the data experts to define the meaning and specify the format of the data type as well
as conducting Data Exploration through various tools in the form of visual information in order
to check for outliers or incomplete information. Data scientists can then determine the appropriate
data cleansing method for data preparation at the next step.
• Data Preparation: It is the process by which a data scientist refines a dataset into a format ap-
propriate, for data mining techniques, by eliminating incomplete data, replenishing missing data
or combining datasets from multiple sources. At this stage, the data scientist selects the data able
to be used for processing the result (Feature Selection). And through the process of understanding
the data, data scientists have to choose a method for cleaning the data, such as filling in incomplete
data or eliminating empty data as well as modifying the types of data ready for data analysis with
data mining techniques such as changing letters into numbers. Then the data scientists will create
a data set from the data that has been cleaned to become a ready-to-use dataset for data analysis
with data mining techniques.
• Modeling: Data scientists choose a data mining technique, Polynomial Regression or Simple
Linear Regression, corresponding to the problem set, to get the results that are in line with the
needs of business people. The results of the analysis of the data that are closest to the facts can
be selected. This is reflected in the processing efficiency of each data mining technique. Once
Supervised-Learning data mining technique has been selected, the data scientists need to consider
data segmentation in order to teach machine learning the Trian Set. Then, the proportion of the
data used in the test was considered (Test Set) before bringing all the data to analyze within the
model. In this step, the data scientist can go back and refine the data to be ready for data analysis
so that the obtained results are more accurate.
• Model Evaluation: Data scientists can select a number of data mining techniques with data analy-
sis capabilities that can be used to serve the same purpose. In other words, to evaluate the model
performance, precision of the model is not the only thing to be considered, but also the consis-
tency of the analysis results. In many cases, the data scientists may find that the results obtained
from the data analysis may not be consistent with the business questions. The scientists can set the
success of data analysis in the business understanding process as a measure of the effectiveness
of data analysis.
• Deployment: This is to take the results to answer business questions in order to support business
decisions. The results can forecast the amount of raw material that should be prepared for the 4th
quarter. In this process, the results obtained from the data mining techniques are numerical. This
requires the interpretation of results in a format that the general public can understand. Therefore,
the data scientists are required to prepare an oral presentation along with the preparation of sup-
porting documents so that business questioners can understand the results obtained from the anal-
ysis. Also, during the presentation the data scientists can pose questions from other perspectives.
Moreover, in future data analysis projects including providing feedback to the organization on the
information, the organization needs to collect and gather data beneficial for future analysis.

5
 
Introduction to Data Mining

From Knowledge Discovery Process and Cross-industry Standard Process for Data Mining: CRISP-
DM. This book summarizes the data analysis process with data mining techniques as shown in Figure 1.

Figure 1. Data analysis using data mining techniques

6

Introduction to Data Mining

As shown in Figure 1, the process starts with understanding business problems and determining
the problems with the management team. Then the survey is conducted and the data are prepared by
considering the resources in order to obtain the appropriate answers to the questions. In this step, data
scientists need to consult with data experts, especially those who are familiar with the industry. Then the
data are prepared for processing with data mining techniques. During processing, the data scientists have
to choose a data mining technique, consistent with their answers. Multiple techniques can be selected
at the same time and the degree of accuracy of each model is tested to obtain the results. Afterward,
the effectiveness of the results is tested by the data experts to reflect whether the results can be used to
answer the questions. In the last section, the data scientists make an oral presentation to interpret the
analysis results to the executives who are also the questioners.

SUPERVISED LEARNING

Regarding Supervised Learning Data Mining Techniques (Abdul and Starkey, 2020), data scientists need
to define the target variables (Data Labelled) used for data analysis. Then the data set is divided into 2
parts. The first part is for teaching machines to learn. The second part is for testing data sets according
to established data mining techniques. The data mining techniques that are classified as Supervised
Learning are as follows.

Classification

Classification is to categorize the data (Maslan, Mohamad and Mohd, 2020). For example, if doctors want
to categorize patients with COVID-19 into 3 categories, asymptomatic patients who show no symptoms,
symptomatic patients and severe patients, other variables are required to classify data according to the
classification rules as follows.
If the diagnosis = COVID-19 and symptoms detected = symptomatic and symptom level = severe
condition, then patient status = severe patient.
If the diagnosis = COVID-19 and symptoms detected = symptomatic and symptom level = mild
symptoms, then patient status = symptomatic patient.
If the diagnosis = COVID-19 and symptoms detected = asymptomatic, then patient status = asymp-
tomatic patient.
The classification can only be done only if there is a result of a variable that supports the effect of a
classification, such as diagnosis, symptoms, and level of symptoms which are the variables of COVID-19.
And the classification outcome is the target variable. Classification rules can be described as Decision
Tree (Romero et. al., 2019) to illustrate variables as presented in Figure 2.
Besides classification rules, the nearest neighbor matching method can also be applied (Atanasovski,
Kostov, Arapinoski and Spirovski, 2020) by identifying what kind of data it is classified from the nearest
information. For example, if the car model data is close to the data of other luxury cars, such car model
will be classified as a luxury car.

7
 
Introduction to Data Mining

Figure 2. Decision tree diagram

Numerical Prediction

Numerical Prediction is classified as Supervise Learning (Liu et. al., 2021) which requires setting the
target variable and analyzing the data from other variables. For example, it can be applied to analyze
students’ height with Simple Linear Regression. Other well- numerical data forecasting include Neural
Networks that can use multiple variables to predict target outcomes, and Deep Learning which is appro-
priate for data processing with large volumes and non-linearity. Through such processing, the datasets
need to be divided; for machine learning and for testing based on numerical forecasting.

UNSUPERVISED LEARNING

Unsupervised Learning is the techniques in which the data scientists do not need to determine the target
variables nor to divide the datasets into machine learning and testing parts (Shrestha, Razavi and Prasad,
2020). All data can be processed to achieve the desired results. The data mining techniques classified
as Unsupervised Learning are as follows:

8

Introduction to Data Mining

Association Rule Mining

Association Rule Mining is used to analyze the relationship of data within a single variable or multiple
variables. For example, marketers need to know what Hashtags they should write to present their prod-
ucts on Twitter to match the needs of customers. In response to such a question, the data scientists can
use Association Rule Mining techniques to analyze the association rules of these data, as presented in
the example (Tanantong and Ramjan, 2021).

Hashtag 1, Hashtag 2, Hastag3, …, Hashtag n => {Supply} or

Hashtag 1, Hashtag 2, Hastag3, …, Hashtag n => {Demand} or
Hashtag 1, Hashtag 2, Hastag3, …, Hashtag n => {Others}

According to the samples, data scientists can perform an analysis of hashtags within Twitter. This
allows marketers to determine which hashtags they should use in order to be consistent with the purchas-
ing needs of their customers within Twitter. For example, #SellingBlackpinkConcertTickets, which is a
hashtag about selling products, together with #LookingForBlackpinkConcertTickets, which is a Hashtag
about the need to purchase products will allow customers to find a group of such Twitter messages faster.

Clustering

Clustering is to group data from 2 or more variables (Reddy and Aggarwal, 2014). Data scientists can
take a series of data on the width and length of the pants legs that former customers bought from the
factory. Then the number of desired groups is specified as presented in the Figure 3.

Figure 3. Clustering of pant sizes that the factory wants to produce

Source: Tanantong and Ramjan (2022)

9
 
Introduction to Data Mining

From the figure, the data scientists can take the data on the pant leg width displayed on the x-axis
and the pant leg length displayed on the y-axis to create a data point, and the data are then divided into
2 groups. The red color represents Group 1, and the blue represents Group 2. In each group, there is an
intersection (Centroid) which the data scientists can use in order to determine the 2 sizes of the pants
that the factory should produce.

DATA ANALYSIS TOOLS

In data science, there are tools to analyze huge amounts of data. The steps of the data analysis and data
collection are as follows:
Data Collection: In the case that the data scientists are in the organization and required to analyze
data, they can access to data sources directly with permission from their organization. But for the sci-
entists who need to use public information such as information from websites or social networks, they
can use tools such as the Twitter API for extracting data from Twitter (Hernandez-Suarez et. al., 2018)
and Python Web Scraping for extracting data from websites (Nigam. et. al. 2021).
Survey and Data Preparation: If the data scientists develop themselves from people working in
the digital field, they can write programs, such as R or Python to explore data. It can develop data vi-
sualizations in various ways. The highlight of these languages is that it is an open-source software so
the purchase of the software is free of no charge. On the other hand, the data scientists, who once was
data experts, may choose to use software packages such as Microsoft Power BI, Tableau or RapidMiner
to explore data. Data scientists can then choose to prepare the data through Python by improving the
data format, combining data from multiple datasets, removing incomplete data, filling incomplete data
or preparing data using models within RapidMiner software. To prepare uncomplicated data, the data
scientists can also use Microsoft Excel software.
Data Processing With Data Mining Techniques: Data scientists can still choose to write a program
with Python in order to process data from libraries. Python has provided for both supervised learning
and unsupervised learning. RapidMiner software can be applied to support modeling in all forms of
data mining processing techniques and displaying both statistical data and visual information. Therefore,
RapidMiner software has gained popularity and overcomes the limitations of the data scientists who lack
programming skills. It assists them to use data mining techniques in their data science work.
Data Presentation: In this process, the data scientists can seek advice from a data expert in order to
verify the effectiveness of the processing. The results shown via Python or RapidMiner can be used to be
relay to the questioner. The data scientists need to translate the results to infer the answers the question-
ers need. In other cases, if such questions do not require the use of data mining techniques for analysis,
the data scientists can also use a variety of software to develop visual data to answer those questions.
Thus, RapidMiner software is a data science tool that supports every step of the data scientist’s work.
There are also machines used for processing with complete data mining techniques, and data scientists
do not need to write programs. This book is, therefore, use RapidMiner software to demonstrate data
processing with various data mining techniques.

10

Introduction to Data Mining

CASE STUDY

World-wide educational institutions have recognized the importance of data science, from the survey
process, data preparation, data visualization and data mining processing. Not only institutions of higher
education, but private-sector academies focus on providing short-term coursework to enable learners to
re-skill, transforming themselves to become data scientists. For those who are data scientists can also
increase their knowledge and abilities even higher. At present, many universities are using online learn-
ing and teaching methods, both in the form of the university itself and partnerships such as Coursera
or EdX. Students can study one subject at a time and receive a certificate. When students complete the
compulsory courses, those certificates can be converted into credits to university coursework in order to
obtain a degree. There are also data scientists who intend to share their knowledge independently from
university teaching. Those scientists can organize their own training courses through online platforms
such as Udemy.
An interesting difference between the teaching and learning management of universities, the private
sectors and independent data scientists is the cost of education. If students are looking for a certificate or
a degree from a world-leading university, they may choose to study through an online platform that the
university manages itself or through a partner platform. To study in this manner, students need to have
a reasonable budget for studying. On the other hand, if students would like to learn without expecting
a certificate from a world-class university, they can save money by choosing to study with independent
data scientists who provide online courses through online platforms. Surely, on the internet, students still
have access to data science resources where data scientists share their passions, a wealth of knowledge
and experience through YouTube channels.
Therefore, this book introduces interesting courses on self-improvement in data science, especially
the courses that are taught through the Online Platform.

UNIVERSITY COURSES TAUGHT THROUGH ONLINE PLATFORM

Nowadays, universities world-wide have become aware of distance learning management. They have
developed both graduated and non-graduated courses for students to study through online platforms
from anywhere and at any time in the world. Besides, in the countries where English is not the official
language, there are also universities that provide courses in their native language having English subtitles
for lectures. Therefore, learners are no longer limited in their educational opportunities by language
barriers. And what is interesting is the First Mover Higher Education Institutions, leading universities
of the world, are stepping into the distance education industry. Regardless of where in the world the
learners reside, they can study at an internationally standard tertiary institution. Therefore, this section
of the book introduces online master’s degree programs related to data science work in which readers
can start studying a series of courses and collect all the credits.
The Master of Computer Science program offered by the University of Illinois provides a Data Science
major in which students gain knowledge of machine learning, data mining, visualization data creation and
processing big data with Cloud Server. Students can study courses to earn credits and gradually combine
them into a Master of Computer Science for graduation, such as the course of Deep Learning Visuals.

11
 
Introduction to Data Mining

Figure 4. Master of Computer Science from University of Illinois

Source: https://www.coursera.org/degrees/master-of-computer-science-illinois

When there is a course improvement, it may result in the students being unable to bring the credits
to combine for a Master of Computer Science graduation, but the students can still complete the afore-
mentioned subjects to enhance their knowledge. However, for those who are interested in studying for
a degree, the University of Illinois offers the following specializations:

• Machine Learning: Is a subject on using machine learning to analyze audio data, geography,
computer vision, and natural language processing.
• Data Visualization: is about creating visual data using Tableau program so that data scientists
can use the developed visual data for effective presentation. The visualization data can also inter-
act to show the details of the data analysis in different dimensions.
• Data Mining: Is regarding the analysis of unstructured data and patterns of the data being
analyzed.
• Cloud Computing: Is about Cloud computing technology, infrastructure development and ap-
plication development to prepare a Cloud environment suitable for analyzing data from Big Data.

When graduating, the learner will acquire the specific skills which are the ability to process big data
in Cloud environment, data analysis skills to support decision-making and principles of using data sci-
ence for research as well as the use of tools to support data analysis with data science.
Such courses are a broad framework for learning content. If students would like to earn a degree, they
will still have to take additional compulsory subjects such as Cloud Computing Capstone and Practical
Statistical Learning, and will also have to take other elective subjects such as Internet of Things and
Methods for Applied Statistics. In all subjects, students must obtain a grade of at least B-level. Students
can also complete all course in 1 year or take up to 5 years. Throughout the course, they can connect
with teachers and teaching assistants through the Discussion Board.
Eligible applicants to study for a degree must have completed a bachelor’s degree, obtaining a minimum
3.20 GPA out of 4.00 and should have a background in object-oriented programming, Data Builder and

12

Introduction to Data Mining

Algorithm. However, students who do not have such skills can enroll for these courses on programming
fundamentals and data structures. And for those who are from Non-Native English-speaking countries
are required to present TOEFL/IELTS results to apply for the course.

Master of Science in Analytics From Georgia Institute of Technology

Figure 5. Master of Science in Analytics, Georgia Institute of Technology Teaching through EdX’s online
platform
Source: https://www.edx.org/masters/online-master-science-analytics-georgia-tech

The Georgia Institute of Technology’s Master of Science in Analytics is a joint program of Georgia
Tech’s College of Engineering, College of Computing, and Scheller College of Business, with majors
focused on three areas: Analytical Tools, Business Analytics and Computation Data Analytics. Students
will also be able to choose from the Analytics: Essential Tools & Methods MicroMasters Program to
study without expecting a degree. The courses can also be credited toward postgraduate study. It consists
of the following courses:

• Introduction to Analytics Modeling: Is a related course to understand data, data forecasting and
data analysis for business decision making. In this subject, students will understand how to choose
the right datasets, algorithms, techniques and patterns in data analysis to support decision-making
in solving business problems. Students will learn R language as an analysis tool and practice mod-
eling in data analysis which covers Classification, Clustering, Change Detection, Data Smoothing,
Validation, Prediction, and Optimization.
• Computing for Data Analysis: Is a course that covers data collection, data storage, data analysis
and visualization data. In this subject, students will use data analysis tools including Python, R,
and SQL, using the software editor, Jupyter notebooks. Students will gain knowledge in quantita-
tive data analysis that integrates statistical data analysis methods with computer programming by
applying those statistics as data processing algorithms.
• Data Analytics for Business Course: Provides students with an understanding of real-world
business case studies and the application of data analytics techniques to support decision-making
to address problems in those case studies. It covers the collection of data, analysis of various forms

13
 
Introduction to Data Mining

of data, exploring patterns of data, bringing the dataset to test according to the hypothesis set. This
course also covers the challenges regarding data analytics the business world is facing. In this
course, R language is used as a tool to practice experimenting with datasets.

Once students have studied in the Micro Master’s course, they can continue to study at the Georgia
Institute of Technology through the university’s Online Platform.

Figure 6. The Georgia Institute of Technology’s Master of Science in Analytics program is offered through
the university’s online platform
Source: https://pe.gatech.edu/degrees/analytics

In order to earn a degree, students must choose to study major subjects, which are divided into 3 majors:

MAJOR: ANALYTICS TOOL

Section 1: Compulsory Courses

• Introduction for Computing for Data Analysis: In this course, students will learn about data
analysis, programming, database access, multidimensional array, quantitative data processing and
the visual data development from case studies and practices.
• Introduction to Analytics Modeling: This course discusses various models and data analysis
techniques including Classification, Clustering, Time Series, regression, experimental design for
testing data distribution, composition analysis, modeling, preparing Model Validation Data, and
selecting features used in Data Analysis.
• Business Fundamentals for Analytics: This course is regarding fundamentals of management
and business terminology, which provide students with an understanding of the concepts enable
businesspeople to work in the business world

14

Introduction to Data Mining

Section 2: Advanced Compulsory Courses

• Data and Visual Analytics: In this course, students will learn how to create multidimensional
visual data by focusing on both theoretical and practical studies with datasets in case studies.
• Data Analytics in Business: This course deals with extracting data from business datasets. It cov-
ers methodology, process V, and challenges related to data analysis in business.

Section 3: Elective Courses in Statistics – Students Can

Choose Two Subjects From the Entire List of Courses

• Time Series Analysis: Forecasting data from stored data based on time in each period.
• Regression Analysis: This course covers the use of statistics for forecasting data by using Bayesian
techniques for data analysis and testing hypotheses.
• Data Mining and Statistical Learning: This course covers the use of data mining techniques
and statistical principles in data analysis such as K-NN, regression analysis, and neural networks,
using R language as a tool for experimenting with data mining techniques.

Section 4: Elective Courses in Practice – Students

Have to Choose One Subject

• Simulation: It covers the use of models to analyze data for problem solving.
• Deterministic Optimization: In this course, students will learn about the concepts of linear al-
gebra and calculus, as well as the concept of calculating suitable values using various techniques.

Section 5: Elective Courses – Students Have to Choose Three Subjects

In this section, students can choose to study subjects in other sections so that the coursework in this section
can complete collecting credits. However, there are other subjects in this section for students to choose.

• Big Data Analytics: In the Health Industry: Students will learn about the analysis and develop-
ment of multidimensional visual data, both theory and practice from case studies in the health
industry.
• Database System Concepts and Design: Students will learn about relational databases, database
design and query processing.
• Machine Learning for Trading: In this course, students will learn about the challenges of apply-
ing Machine Learning to Trading, algorithms for collecting data from market orders, data analysis
for trading decisions using Linear regression, KNN and Regression Trees that are appropriate for
each case.

Section 6: Applied Analytics Practicum

In this section, students need to develop projects in the Analytics Tool by bringing business problems
from the real cases of their work to use as topics in project preparation. Students are required to produce

15
 
Introduction to Data Mining

a report explaining the scope of the project, data analysis purposes, project management plan and results
in data analysis.

MAJOR: BUSINESS ANALYTICS

Section 1: Compulsory Courses

• Introduction for Computing for Data Analysis: In this course, students will learn about data
analysis, programming, database access, multidimensional array, quantitative data processing and
the development of visual data from case studies and practice.
• Introduction to Analytics Modeling: This course discusses various data analysis models and
techniques, including Classification, Clustering, Time Series, regression, experimental design for
testing the data distribution, composition analysis, modeling, preparing Model Validation data
and selecting Features to use in data analysis.
• Business Fundamentals for Analytics: This course deals with fundamentals of management and
business terminology to provide students with conceptual understandings enable business people
to work in the business world.

Section 2: Advanced Compulsory Courses

• Data and Visual Analytics: In this course, students will learn how to create multidimensional
visual data by focusing on both theoretical and practical studies with datasets in the case studies.
• Data Analytics in Business: This course deals with extracting data from business datasets. It cov-
ers methodology, process V, and challenges related to data analysis in business.

Section 3: Elective Courses in Statistics – Students Choose

Two Subjects From the Entire List of Courses

• Time Series Analysis: Forecasting data from stored data based depending on time in each period.
• Regression Analysis: This course covers the use of statistics to forecast data by using Bayesian
techniques for data analysis and hypotheses test.
• Data Mining and Statistical Learning: This course covers the use of data mining techniques and
statistical principles in data analysis such as K-NN, regression analysis, and neural networks using
R language as a tool for experimenting with data mining techniques.

Section 4: Elective Courses in Practice – Students

Have to Choose One Subject

• Simulation: This course covers the use of models to analyze data in problem solving.
• Deterministic Optimization: In this course, students will learn about the concepts of linear alge-
bra and Calculus, as well as the concept of calculating suitable values using various techniques.

16

Introduction to Data Mining

Section 5: Elective Courses – Students Need to Choose Three Subjects

In this section, students can choose to study subjects in other sections so that the coursework in this section
can complete collecting credits. However, there are other subjects in this section for students to choose.

• Digital Marketing: Students will explore a wide range of digital marketing concepts including
mobile marketing, email marketing and social media marketing, and the use of marketing analyt-
ics tools such as search engine optimization, pay-per-click advertising.
• Financial Modeling: In this course, students will learn about creating spreadsheets using pivot
tables, using Excel functions to analyze data, writing VBA. It covers applications in analysis, fi-
nancial report, cost analysis and pricing. Students therefore can build financial models.

Section 6: Applied Analytics Practicum

In this section, students need to develop projects in the Analytics Tool by bringing business problems
from the real cases of their work to use as topics in project preparation. Students are required to produce
a report explaining the scope of the project, data analysis purposes, project management plan and results
in data analysis.

MAJOR: COMPUTATIONAL DATA ANALYTICS

Section 1: Compulsory Courses

• Introduction for Computing for Data Analysis: In this course, students will learn about data
analysis, programming, database access, multidimensional array, quantitative data processing,
and the development of visual data from case studies and practices.
• Introduction to Analytics Modeling: This course discusses various data analysis models and
techniques including Classification, Clustering, Time Series, regression, experimental design for
testing the distribution of data, composition analysis, modeling, Preparing Model Validation Data,
and Selecting Feature used in Data Analysis.
• Business Fundamentals for Analytics: This course deals with fundamentals of management and
business terminology to provide students with conceptual understanding enable business people
to work in the business world.

Section 2: Advanced Compulsory Courses

• Data and Visual Analytics: In this course, students will learn how to create multidimensional
visual data, focusing on both theoretical and practical studies with datasets in the case studies.
• Data Analytics in Business: This course deals with extracting data from business datasets. It cov-
ers methodology, process V, and challenges related to data analysis in business.

17
 
Introduction to Data Mining

Section 3: Elective Courses in Statistics – Students Choose

Two Subjects From the Entire List of Courses

• Time Series Analysis: This course deals with data forecast from stored data based on time in each
period.
• Regression Analysis: This course covers the use of statistics for data forecast, using Bayesian
techniques for data analysis and testing the hypotheses.
• Data Mining and Statistical Learning: This course covers the use of data mining techniques
and statistical principles in data analysis such as K-NN, regression analysis, and neural networks,
using R language as a tool for experimenting with data mining techniques.

Section 4: Elective Courses in Practice – Students

Have to Choose One Subject

• Simulation: This course covers the use of models to analyze data for problem solving.
• Deterministic Optimization: In this course, students will learn about the concepts of linear alge-
bra and Calculus, as well as the concept of calculating suitable values using various techniques.

Section 5: Elective Courses – Students Need to Choose Three Subjects

In this section, students can choose to study subjects in other categories so that the coursework in this sec-
tion can collect complete credits. However, there are other subjects in this section for students to choose.

• Knowledge-Based AI
• Reinforcement Learning
• Deep Learning

Section 6: Applied Analytics Practicum

In this section, students need to develop projects in the Analytics Tool by bringing business problems
from the real cases of their work to use as topics in project preparation. Students are required to produce
a report explaining the scope of the project, data analysis purposes, project management plan and results
in data analysis.
With a wide variety of majors, learners can choose from categories and majors for flexibility in
graduation. Georgia Institute of Technology guarantees the quality of education through EdX’s online
platform, provided by the Georgia Institute of Technology itself, as it is equivalent to studying on campus
by equipping the same faculty teaching onsite at the Georgia Institute of Technology. When students
enroll the Master of Science in Analytics course, they receive an orientation from the course director to
be well prepared and enabled to write through Online Platform until graduation. The cost of studying
is approximately 275 USD per credit, so throughout the course students can budget for approximately
9,900 USD with additional tuition fees of 301 USD per semester.

18

Introduction to Data Mining

Figure 7. Master of Science in Digital Business Transformation (Data Science) from Thammasat University
Source: https://www.skilllane.com/academic/tuxsa/datascience

A leading university named Thammasat University is located in a Southeast Asia, Thailand. The uni-
versity has provided a course in the Master of Science in Digital Business Transformation (Data Science)
course. This course is a collaboration between Thammasat University and Online Platform Skilllane,
which is an educational Start Up in Thailand in organizing an online master’s degree program under
the name TUXSA in Thai. TUXSA also means skill, corresponding to SkillLane’s name. The Master of
Science in Digital Business Transformation (Data Science) program is under the College of Innovation,
one of the faculties of Thammasat University.
In the aforementioned course, various learning styles are provided to grant certificates for each course.
Students can choose to study to improve their knowledge or choose to study to collect credits for further
study to receive a degree at Thammasat University by registering for the final exam after completing
the course content. They can also study to receive a Data Science certificate after gaining 18 credits.
Compulsory courses are as follows.

• Python for Data Science: In this course, students will learn about the fundamentals of program-
ming using Python, problem analysis, Algorithm of sequential programming, programming for
creating conditions, programming for reprogramming the computer, and case studies in the field
of data science.

19
 
Introduction to Data Mining

• Python for Advanced Data Science Applications: Students will explore the relationship of
Python programming to data science, the structure of Supervised-Learning and Unsupervised-
Learning, the use of data mining techniques in data analysis, and visual data presentation.
• Business Intelligence: In this course, students will learn about the principles of integrating data
from various sources, designing indicators to measure business performance, data extraction, fea-
ture engineering, dashboard designing, and the use of Microsoft Power BI as practical tools for
case study practices.
• Text Analytics: This is a course on data mining techniques used to extract text patterns and
data from websites, evaluation of the efficiency of the algorithm used to extract the data, Semi-
structured and unstructured data relationships for data drilling and limitations of data mining in
each model, application of forecasting and data clustering techniques, and evaluating the effec-
tiveness of the results obtained from data drilling.
• Data Mining Algorithm: In this course, students will learn about the concept of accessing data
within a database and data mining, data extraction from data mining, the use of Decision Tree,
classification and clustering, and current data Mining case studies.
• Advanced Data Mining Algorithms: This course builds on Data Mining Algorithms, aiming
to provide learners with essential data mining techniques including Neural Network, K Nearest
Neighbor (K-NN) and Logistic Regression, along with applied case studies, using data mining in
different industries.
• Data Exploration: Students will learn about the process of exploring data, access to resources,
understanding data types and learning about exploring data techniques using R language as an
experimental tool along with case studies.
• Data Preprocessing: Students will realize the importance of data preparation before using it in
data mining analysis, covering data selection, data cleaning, data integration, data reduction, and
data engineering.
• Fundamental of Digital Transformation: Students will learn about creating digital competence
for organizations, designing customer experiences with digital technology, and internal process
improvement.
• Advanced Digital Transformation: It is a course about modifying business models, change man-
agement, engaging people in the organization for change, instilling digital skills for people in the
organization, and digital platforms that affect internal processes of the organization.
• Data Science Capstone Project: It is a data science project consisting of questioning for data
analysis, seeking for data, data mining, data preparation, using models for data analysis, data vi-
sualization, and data interpretation.
• Data Science Research Methodology: This course prepares students for independent research
consisting of Research problems, hypotheses, literature reviews, quantitative data analysis, quali-
tative data analysis, experimental data analysis, hypothesis testing, interpretation of results, sum-
marizing results, citations, and writing research reports and academic articles.

When students have completed all the courses, they can apply for the Comprehensive Examination
and the result is used for admission to the College of Innovation Thammasat University. When becoming
students at Thammasat University, they can also take elective courses and conduct independent research
to complete their studies. The elective courses are as follows:

20

Introduction to Data Mining

• Business Case for Data Science: This is a course that discusses the context and the use of data
across industries and within organizations, adaptation of the organization to development of data
science today, the importance of different data types influencing the growth of a business or orga-
nization, the role of data analysts and the challenges and barriers of data science, future trends in
data science, the relationship of data and artificial intelligence, and application of artificial intel-
ligence in business.
• Developing Solution and Engagement: Students will learn how to apply digital technology to
solve problems at organizational and national levels by searching for connections between prob-
lems and types of technology in order to analyze, plan, create or use technology to meet the needs
and solve problems at organizational and national levels with case studies.
• Cyber Security Policy: This course includes contents in Cybersecurity Technology, Encryption,
Economics of Cybersecurity, Cyber Security Policy both nationally and internationally, Critical
Infrastructure Protection, Disaster Recovery Planning, and Cybersecurity case studies.
• Cyber Threat and Security: This course covers internet technology and networking, information
technology system security, Internet threats at the malware level, cyber theft, movement for the
benefit of the group and cyber warfare, and Privacy and Cyber Surveillance.
• Digital Leadership: In this course, students will be prepared to become executives in digital orga-
nizations by studying about communication skills, human relations management, conflict resolu-
tion, effects of leadership on Team Performance, morality, ethics and leadership, social conduct,
environment, and human rights.
• Digital Organization Behavior: This course focuses on managing organizations that use digital
technology, regarding leadership in the digital age, principles of Agile Management, leadership
and risk management, awareness of innovation and creativity, incubating innovation, corporate
culture, and optimization in the organization.
• Design Thinking for Business Strategy and Entrepreneurship: This course includes contents
in design thinking process for developing a successful digital business strategy, and bringing the
design thinking process to the opportunity. It starts with identifying stakeholder needs and barri-
ers to create technological advantages that drive business growth, and to develop a digital product
strategy to be in line with the market and operating system.
• Design Thinking for Digital Society Architecture: It is to apply design thinking processes to
design social architecture in the digital era in order to achieve holistic harmony, and to apply the
results of the design thinking process to analyze the social overview as well as managing differ-
ences in environment and culture that arise from the rapid change through the drive of digital
technology.

The curriculum is 100% online where students can do their independent research and contact advisors
to ask for advice, and do Defense Exam through online channels. Innovative College staff will schedule
exams that are suitable for both the students and the committee.
Regarding qualifications of the applicants, those who have not completed the bachelor’s degree can
study the aforementioned subjects. However, the students are required to have a bachelor’s degree in
order to become an official Master degree student. Therefore, for those who are currently studying at
high school or bachelor’s level, can study various subjects and collect credits beforehand. And once
having a bachelor’s degree, they can then apply all the qualifications and compulsory courses that have
been studied to register a Master degree here.

21
 
Introduction to Data Mining

Figure 8. Data and Policy short-term course hosted by The University of Chicago
Source: https://info.harris.uchicago.edu/summer-scholar-program?hsLang=en

This short course from Harris School of Public Policy, The University of Chicago is delivered
through the University’s Platform by bringing the Domain of Public Policy as a case study throughout
the training. It covers data collection, data cleaning, data analysis and data visualization development
to solve the social problems. The program is also suitable for those with no experience in data science,
as well as students or professionals who would like to re-skill to become the data scientist profession.
The course covers the following contents.

• Principles of Statistical Data Analysis: It deals about the use of data analysis tools, a data
analysis process that supports the development of public policy, the use of case studies to analyze
problems, data analysis method, and interpretation of the results obtained from the data analysis.
• Basic R Programming Language: It is suitable for those who have no programming background.
It enables learners to enter the basics of programming with R language, data extraction, data
cleaning, data analysis and data visualization by using R language to write command sets.
• Capstone Research Project: The course prepares teachers to supervise in researches and cre-
ate networking among students in a project. The project covers the process of research design
and public policy analysis. The process is also consistent with the data analysis methods of data
science work; data collection, data analysis, data visualization and model policy creation. The
trainees will be able to apply the knowledge from the project to analyze the data in other domains.

After completing this course, students will receive an Official UChicago Transcript containing grades
for each subject and a Certificate of Credential Completion for each trained subject.

22

Introduction to Data Mining

In this course, students can access the videos of each lesson and each course over and over again
through The University of Chicago’s Platform. Students can also request a consultation with a course
assistant through the Q&A. Regarding the Capstone Research Project, students will also have the oppor-
tunity to meet and consult with teachers in Live Lectures as well as work in groups with other learners
who are attending the same class during the same period.
Regarding qualifications, students must be at least 18 years of age and have completed the first year
of a bachelor’s degree. Due to the fact that students have to work in groups with others, this course is
conducted several times a year. Students can apply for classes according to the cycles. In this course
some scholarships are also provided for the applicants who are eligible for scholarships.
Universities worldwide are keen to provide distance learning programs for both degree programs
and non-degree programs. And after the COVID-19 crisis, these universities have adapted to teaching
in the form of Live Sessions where learners have to access the class via applications. From such events,
learners all over the world are acknowledging the effectiveness of video-based learning that is recorded
on various online platforms. Thus, the direction of campus education has become a stepping stone in
management evolution for university education.

COURSES OF THE INDUSTRY THAT ARE TAUGHT

THROUGH ONLINE PLATFORMS

Industry around the world is experiencing a situation where the graduates or those who have graduated
from short courses cannot work or do not have skills that meet the needs of that industry. In the data sci-
ence industry, the digital data leading companies have created their own courses to produce employees
to meet their needs. Those who have graduated from those courses can also use the certificate to apply
for a job with any company that provides teaching and learning in data science on an online platform.
This book, therefore, provides examples of data science courses provided by the world’s largest
companies that would like to create skilled workers in line with their operational needs.

IBM Data Science Professional Certificate

IBM is a private company who produces Online Courses, especially in Professional Certificate, in col-
laboration with Coursera, an online platform in education. The tutors are IBM’s data science experts.
Learners can choose to earn a certificate for each course of a total of 10 courses, and combine all 10
courses to earn an IBM Data Science Professional Certificate.
The IBM Data Science Professional Certificate aims to help learners understand the concepts of
working in data science, developing skills in using data analysis tools, data cleaning, creating visual
data and using machine learning for data analysis focusing on the application of Python. This course
is also suitable for those with no computer science background or programming skills. Similar to other
short-term courses, this course promotes Re-Skill for those who want to transition from other careers to
becoming a data scientist. The course consists of the following subjects.

• Comprehension of Data Science Work: It discusses the concept of data processing with an
ancient Egyptian case study that ties taxation to the amount of water in the Nile to reflect the link
between past processing and the value of Processing with today’s data science principles.

23
 
Introduction to Data Mining

Figure 9. Certificate in Data Science by IBM

Source: https://www.coursera.org/professional-certificates/ibm-data-science

• Well-Known Data Science Tools: There are Jupyter Notebooks, RStudio and GitHub, Computer
programming in data science using Python and R. In this course, students also create projects
related to data science.
• Process of Data Science Work: It discusses how to identify and analyze problems from business
case studies including data collection, data analysis, modeling and understanding user feedback
after the model is implemented to reflect the effectiveness of the model, which leads to solutions
to specific business problems.
• Use of Python in Data Science and Artificial Intelligence: It starts with the basics in program-
ming, data structure, data analysis with practice in programming as well as making projects based
on the knowledge learned in this course. Python classes in Data Science and Artificial Intelligence
can also be integrated into other IBM courses, including the IBM Applied AI Professional
Certificate, Applied Data Science Specialization, and IBM Data Science Professional Certificate.
• Creating a Python Project for Data Science: This is to bring the data set, a case study, to dem-
onstrate for students in writing a program to analyze data in the form of Dashboard, and when
students want to register for this course, they need to pass the courses on the use of Python in data
science and artificial intelligence before.
• Database and SQL for Data Science With Python: This course discusses the concept of
Relational Database, retrieving data from a database by writing SQL statements, building a data-
base in the cloud, and applying SQL statements with Python from Jupyter Notebooks.
• Python Data Analysis: Students will learn how to use Python for data exploration, data prepara-
tion before data analysis, the use of simple statistics to analyze data, preparation of visual data,
data forcast, and data analysis models which covers libraries including Pandas, Numpy and Scipy,
Scikit-Lern.
• Data Visualization With Python: This course applies storytelling to communication with data
visualizations, Python programming for data extraction, understanding the data and the use of
such data to support decision, which covers the Matplotlib, Seaborn and Folium libraries.

24

Introduction to Data Mining

• Machine Learning With Python: This subject discusses a case study of real-world application
of Machine Learning, principles of Supervised-Learning and Unsupervised-Learning, Evaluation
of model performance, Algorithms of Machine Learning Processing. This course discusses data
mining techniques including Regression, Classification and Clustering. The case studies are can-
cer detection, forecasting economic trends, forecasting the likelihood that users will switch to
other product services, and referral systems.
• Applied Data Science Capstone: Students will assume the role of a data scientist for a startup
company. Students will use the data science processes to analyze data. It includes surveying data,
data visualization, data mining model development, evaluation of model performance, and inter-
preting the analysis results for the company so that the company can take that information into
account for operating costs. The course will prepare a data set for learners to analyze.

Google Analytics Professional Certificate

Figure 10. Google Analytics Professional Certificate

Source: https://www.coursera.org/professional-certificates/google-data-analytics

The Google Analytics Professional Certificate course by Google, in partnership with Coursera Online
Platform, focuses on building skills for anyone without restrictions on educational qualifications. Learn-
ers with no background and degrees in data science can attend such courses. This course covers data
collection, data processing, data compiling for presentation and the use of data from the processing to
make decisions. The course instructors are data science experts working at Google. Students will also be
able to share hands-on experiences from those involved in the data science industry. Upon completion
of the course, students can also get a certificate for applying to work with Google companies.

• Foundations: Data, Data, Everywhere: In this course, students will learn how to use SQL and R
to process data, data science process as well as ecosystem terminology in data science work. This
course will bring real case studies about data analysis to develop new products and improve inter-
nal processes of the organization used for data analysis. Students will gain experiences to become
a Junior Data Scientist. This creates the opportunity to apply for a data scientist job.

25
 
Introduction to Data Mining

• Ask Questions to Make Data-Driven Decisions: This course aims to help learners organize their
thinking for effective questioning, which leads to the pursuit of resources that can be used for data
processing. This course also prepares students to use worksheet software in data processing.
• Prepare Data for Exploration: In this course, students will learn how to determine the data
sources suitable for analysis, structured or unstructured data types, characteristics of the bias types
during data collection which affects the analysis and interpretation of the data processing results,
the use of SQL to retrieve data within a database, data filtering, data compilation and data security.
• Process Dirty for Data to Clean: In this course, Google experts will use real-world case studies
as an experimental tool. Learners will learn how to verify data completeness, data cleaning, the
use of SQL for data cleaning and data processing, comprehension of data level probably used for
processing after data cleanup, and preparation of data cleaning report.
• Analyze Data to Answer Questions: In this course, students will learn about a variety of data
analysis methods Feature Engineering process, the use of formulas and functions within Spread
Sheet Software and data processing using SQL language.
• Share Data Through the Art of Visualization: This course deals with the study of visual data
development, creating a story to explain the visual data, practical use of Tableau for visualization
and filtering, effective presentation, limitations that can arise and effective interaction with the
audience.
• Data Analysis With R Programming: In this course, students will learn how to use Software
Editor, which is R Studio, to write programs with R language, to use commands from the Tidyverse
library, to comprehend the Data Frame and its implementation, and to develop visual data with R
language.
• Google Data Analytics Capstone: Complete a Case Study: In this course, students will explore
case studies in implementing data science processes in real-world datasets, including asking ques-
tions, data analysis, interpretation, visual data development. This course also provides students
with interview questions for job vacancies in the field of data science.

Although students can choose to study any subject first, this course is designed for learners to study in
an organized sequence in order to prepare students for professional skills in data science. The certificate
they have received can be used to apply for a job anywhere, especially at Google company.
It is not a new trend that the industrial sector intends to produce workers that meet their own needs,
so teaching and learning through Online Platform has become a new recruitment method. Moreover,
due to the COVID-19 crisis spreading around the world, employees do not need to go to work at the
company but work from home, so the company can recruit employees to work for them from anywhere.
The potential of employees can be improved by taking the courses that the company provides on the
Online Platform.

COURSES TAUGHT BY INDEPENDENT DATA SCIENTISTS

At present, there are many online platforms, which is one way to generate income for those who are
highly skilled in various subjects and intend to bring their own knowledge to develop into courses for
those who are interested. The emergence of this courses has greatly influenced the concept of non-credit
education. And as a result, the value of graduate education is dwindling. In addition, the importance of

26

Introduction to Data Mining

the educational system through higher education institutions has also been decreasing because learn-
ers can study through the Online Platform with experts in various subjects. There is no limitation in
recruiting and study time is flexible, so student can choose to take only the subjects that they are really
interested. In addition, it is open for professionals to produce their own material, which has also become
an opportunity for learners to choose instructors with high expertise and good transferability. In turns,
the students feel it is worth the tuition fees they have paid.
The data science experts themselves, expectedly data scientists who are not professors in higher educa-
tion institutions, have foreseen such avenues and produced many courses in data science varied from at the
preliminary to higher levels through the online platforms. This book has picked up examples of courses
that independent data scientists have developed and published through the online platform, as follows.

Python for Data Science and Machine Learning Bootcamp

Figure 11. Python for Data Science and Machine Learning Bootcamp via Udemy online platform
Source: https://www.udemy.com/course/python-for-data-science-and-machine-learning-bootcamp/

In Python for Data Science and Machine Learning Bootcamp, which produces courses from Jose Portilla
through Udemy’s online platform, learners will learn about Python programming for data science and
machine learning, and apply machine learning algorithms to data processing. It covers critical libraries
such as Pandas for data analysis, Seaborn and Matplotlib for visualization, Plotly for Interactive Visu-
alizations, NumPy for Quantitative Data Analysis, and Scikit-Learn library for running commands for
data mining techniques including K-Mean Clustering, Logistic Regression, Linear Regression, Random
Forest Decision Trees, Natural Language Processing and Spam Filters, Neural Network and Support
Vector Machines.

27
 
Introduction to Data Mining

In this course, Jupyter is used as a software editor. The instructor also prepares a Capstone Project
on financial data analysis to allow students to learn Python programming in real situations. The final
part of the course also discusses Big Data processing with Spark, so students have a glimpse into the
application of Python programming in Big Data processing.
However, for this subject, students need to have a good background in programming and the basic
commands in Python before they can begin to learn the different types of command sets involved in
processing. In the course, teachers have provided more than 100 Clips of teaching materials with detailed
command sets.

Data Science and Machine Learning Bootcamp With R

Figure 12. Data Science and Machine Learning Bootcamp with R

Source: https://www.udemy.com/course/data-science-and-machine-learning-bootcamp-with-r/

R is considered one of popular computer languages. Data Science and Machine Learning Bootcamp with
R is an online course developed by Jose Portilla via Udemy online platform. Students will learn how to
get started with installing R Studio and the R Translator, then learn about Matrix, Data Frames, Lists,
Fundamentals of R Programming such as structural programming, recursive programming, conditional
programming, visual data development, using machine learning for data analysis within R language,
Linear Regression, Logistic Regression, K Nearest Neighbors, Decision Tree and Random Forest, Sup-
port Vector Machine, K- Mean Clustering, Natural Language Processing, Neural Nets.

28

Introduction to Data Mining

Regarding data mining techniques and the development of visual data, instructors also provide a
Capstone Project to every lesson so that students can experiment with data analysis from the data set
that corresponds to the questions.
Therefore, this course is suitable for those who have no programming experience at all. The teacher
will teach the foundation of programming using the R language as an experimental tool, and the applica-
tion of R language for data analysis through over 100 video clips.

The Ultimate Hands-On Hadoop: Tame Your Big Data

Figure 13. The ultimate hands-on Hadoop: Tame your big data
Source: https://www.udemy.com/course/the-ultimate-hands-on-hadoop-tame-your-big-data/

In The Ultimate Hands-On Hadoop: Tame Your Big Data course, developed by Frank Kane from Sundog
Education Team on the online platform Udemy, students will learn about designing Distributed Systems
for Big Data Management using Hadoop and Data Engineering Technologies, data analysis on Hadoop
Cluster, relational data analysis using Hive and MySQL, non-relation data analysis using HBase, Cassan-
dra and Mong DB, data querying, interactive using Dirll, Phoenix and Presto, choosing the right storage
technology, understanding Hadoop Cluster Management using YARN, Tez, Zookeeper, Zeppelin, Hue
and Oozie, real time Streaming Data management with Kafka, Flume, Spark Streaming, Flink and Storm.

29
 
Introduction to Data Mining

In this course, the instructor provides hand-on lessons with over 14-hours-long video files that will
provide students access to Big Data management on Hadoop. From business case studies, the learners
are required to have a basic knowledge of programming in Python. Therefore, this course is ideal for
Software Engineers and Developers who intend to gain more knowledge in the Hadoop ecosystem and
how to use Hadoop to store and analyze data in Big Data, as well as Project or Product Managers who
intend to understand Hadoop’s architecture. Moreover, Data analyzes or Database administrators can
also apply knowledge from this subject to their work as well.
From the example mentioned above, it can be seen that independent data scientists are experts in
data science from their own hand-on experience and have the same content transfer skills. Therefore, for
those who are already working in the data science industry and studying for a degree, it does not affect
their career development as they can choose to study through this online platform.

KNOWLEDGE RESOURCES ON YOUTUBE CREATED

BY INDEPENDENT DATA SCIENTISTS

From the case studies mentioned above, all of them are paid educational courses. However, within
YouTube there are independent data scientists who are generous to share their knowledge to the public
on a non-profit basis. Materials on data science are produced and uploaded on YouTube for anybody to
learn. This book has picked up examples of teaching materials on YouTube as follows.

Data Analysis With Python: Full Course for Beginners

(Numpy, Pandas, Matplotlib, Seaborn)

The data analysis with Python – Full Course for Beginners (Numpy, Pandas, Matplotlib, Seaborn) is
produced by Santiago Basulto of RMOTR and is 4.22- hour-long, covering the topic of an overview of
Python in data science, demonstration of using the Pandas library, Seaborn and the SQL language within
Python for data analysis, using Jupyter Notebooks as a tool for testing command set, using Numpy for
Data Pre-Processing, using the Pandas library for creating and manipulating data frames, and fetching
external image datasets into commands.
Therefore, this course is suitable for those who intend to learn about data science and learn to write
computer programs and to prepare the data for further data analysis before deciding to continue to more
complex subjects.

TensorFlow 2.0 Complete Course: Python Neural

Network for Beginners Tutorial

In addition to the introductory courses suitable for those who intend to get started with their data science
careers, YouTube also offers free and fixed lessons. This course deals with high-level data analysis tech-
niques such as the TensorFlow 2.0 Complete Course – Python Neural Network for Beginners Tutorial,
which discusses programming in Python for data analysis using the Neural Network technique. This
course was developed by Tim Ruscica.

30

Introduction to Data Mining

Figure 14. Data analysis with Python: Full course for beginners (Numpy, Pandas, Matplotlib, Seaborn)
Source https://www.youtube.com/watch?v=r-uOLxNrNk8&list=RDLVua-CiDNNj30&index=4

Figure 15. TensorFlow 2.0 complete course: Python neural network for beginners tutorial
Source: https://www.youtube.com/watch?v=tPYj3fFJGjk&list=PLWKjhJtqVAblStefaz_YOVpDWqcRScc2s&index=1

31
 
Introduction to Data Mining

In this course, the content covers Principles of Machine Learning, Introduction to TensorFlow, Core
Learning Algorithm, Neural Networks with TensorFlow, Convolutional Neural Networks, RNNs, Re-
inforcement Learning with Q-Learning.
According to the content, this course is suitable for anyone who has a background in Python program-
ming and intend to practice Python programming to apply machine learning. In each lesson, the teacher
explains the steps and commands in detail so that students can apply their knowledge to any other data set.
As is obvious, YouTube has free-of-cost contents available for both beginners and advanced data
science students. Teachers may only expect learners to press Like, Share and Subscribe. And it is a good
start for those who are not financially ready but would like to improve themselves.

CONCLUSION

Industries in the world see benefits from using data mining techniques to respond to business inquiries.
Therefore, it is necessary to provide an environment conducive to data analysis such as creating a data
scientist and purchasing software and hardware to facilitate big data processing. Data mining techniques
are divided into Supervised Learning, which includes numerical classification and forecasting; and Un-
supervised Learning model, which includes analysis of correlation and segmentation rules. RapidMiner
software is popularly used because it can support every step of data analysis where the data scientists
can create models for data preparation and analysis without writing any program.

EXERCISE QUESTIONS

1. Give an example of the application of data mining techniques in various industries.

2. Describe challenges of Scalability on Data Mining Processing.
3. Explain the meaning of Supervised Learning.
4. Explain the meaning of Unsupervised Learning.
5. Explain the differences of data analysis by program writing and using software packages.

REFERENCES

Abdul Aziz, A., & Starkey, A. (2020). Predicting Supervise Machine Learning Performances for Senti-
ment Analysis Using Contextual-Based Approaches. IEEE Access : Practical Innovations, Open Solu-
tions, 8, 17722–17733. doi:10.1109/ACCESS.2019.2958702
Almasoud, A. M., Al-Khalifa, H. S., & Al-Salman, A. S. (2019). Handling Big Data Scalability in Biologi-
cal Domain Using Parallel and Distributed Processing: A Case of Three Biological Semantic Similarity
Measures. BioMed Research International, 2019, 1–20. doi:10.1155/2019/6750296 PMID:30809545

32

Introduction to Data Mining

Atanasovski, M., Kostov, M., Arapinoski, B., & Spirovski, M. (2020). K-Nearest Neighbor Regression
for Forecasting Electricity Demand. 2020 55th International Scientific Conference on Information, Com-
munication and Energy Systems and Technologies (ICEST), Information, Communication and Energy
Systems and Technologies (ICEST), 2020 55th International Scientific Conference On, 110–113.
Bell, J. (2015). Machine learning: hands-on for developers and technical professionals. John Wiley & Sons.
Berman, J. J. (2018). Principles and practice of big data: preparing, sharing, and analyzing complex
information (2nd ed.). Academic Press/Elsevier.
Bramer, M. (2016). Principle of Data Mining (3rd ed.). Springer-Verlag London Ltd.10.1007/978-1-
4471-7307-6
Cardona, J. A. S., Gomez G, C. H., & Lopez Trujillo, M. (2014). Knowledge discovery process in the
open government Colombian model. 2014 9th Computing Colombian Conference (9CCC), Computing
Colombian Conference (9CCC), 2014 9th, 96–102. 10.1007/978-1-4471-7307-6
Chiarini, T. M., Kohli, R., & Forsgren, N. (2021). Theories in Flux: Reimagining Theory Building in
the Age of Machine Learning. Management Information Systems Quarterly, 45(1), 455–459.
Gupta, A., Sahayadhas, A., & Gupta, V. (2020). Proposed Techniques to Design Speed Efficient Data
warehouse Architecture for Fastening Knowledge Discovery Process. 2020 IEEE Third International
Conference on Artificial Intelligence and Knowledge Engineering (AIKE).
Hernandez-Suarez, A., Sanchez-Perez, G., Toscano-Medina, K., Martinez-Hernandez, V., Sanchez,
V., & Perez-Meana, H. (2018). A Web Scraping Methodology for Bypassing Twitter API Restrictions.
Academic Press.
Jiang, S., Jiang, T., & Wang, L. (2020). Secure and Efficient Cloud Data Deduplication with Ownership
Management. IEEE Transactions on Services Computing, 13(6), 1152–1165.
Kyaw, K. S., & Limsiroratana, S. (2019). Case Study: Knowledge Discovery Process using Computation
Intelligence with Feature Selection Approach. 2019 17th International Conference on ICT and Knowledge
Engineering (ICT&KE), ICT and Knowledge.
Lawal, M. M., Ibrahim, H., Sani, N. F. M., & Yaakob, R. (2020). Analyses of Indexing Techniques on
Uncertain Data With High Dimensionality. IEEE Access : Practical Innovations, Open Solutions, 8,
74101–74117. doi:10.1109/ACCESS.2020.2988487
Liu, Y.-Y., Li, L., Liu, Y.-S., Chan, P.-W., Zhang, W.-H., & Zhang, L. (2021). Estimation of precipitation
induced by tropical cyclones based on machine-learning-enhanced analogue identification of numeri-
cal prediction. Meteorological Applications, 28(2). Advance online publication. doi:10.1002/met.1978
Maslan, A., Mohamad, K. M. B., & Mohd Foozy, F. B. (2020). Feature selection for DDoS detection
using classification machine learning techniques. IAES International Journal of Artificial Intelligence,
9(1), 137–145. doi:10.11591/ijai.v9.i1.pp137-145
Nigam, H., Biswas, P., Raj, J. S., Iliyasu, A. M., Bestak, R., & Baig, Z. A. (2021). Web Scraping: From
Tools to Related Legislation and Implementation Using Python. Academic Press.

33
 
Introduction to Data Mining

Romero, M. P., Parry, J., Prosser, A., Upton, P., Rees, E., Tearne, O., Arnold, M., Chang, Y.-M., Brunton,
L. A., Stevens, K., & Drewe, J. A. (2019). Decision tree machine learning applied to bovine tuberculosis
risk factors to aid disease control decision making. Preventive Veterinary Medicine, 175. PMID:31812850
Samasiri, P. (2020). Visual representations of different family structures using an interactive data explo-
ration technique. 2020 1st International Conference on Big Data Analytics and Practices (IBDAP), Big
Data Analytics and Practices (IBDAP), 2020 1st International Conference On, 1–5.
Shrestha, R. B., Razavi, M., & Prasad, P. W. (2020). An Unsupervised Machine Learning Technique
for Recommendation Systems. 2020 5th International Conference on Innovative Technologies in Intel-
ligent Systems and Industrial Applications (CITISIA), Innovative Technologies in Intelligent Systems and
Industrial Applications (CITISIA), 2020 5th International Conference On, 1–9.
SPSS Inc. and International Consortium Publish CRISP-DM 1.0; Free Document Offers Step-By-Step
Guidance for Data Mining Projects. (2000). Business Wire.
Sunkpho, J. & Ramjan, S. (2020). Predicting Condominium Price in Bangkok Using Web Mining Tech-
niques. Srinakharinwirot Research and Development (Journal of Humanities and Social Sciences), 12(24).
Tanantong, T. (2021). Data Mining. Retrieved from: https://www.skilllane.com/courses/tuxsa-Data-
Mining-Algorithms/chapter
Tanantong, T., & Ramjan, S. (2021, April-June). An Association Rule Mining Approach to Discover
Demand and Supply Patterns Based on Thai Social Media Data. International Journal of Knowledge
and Systems Science, 12(2), 1–16. doi:10.4018/IJKSS.2021040101
Tanantong, T., & Ramjan, S. (2022). Data Visualization for Data Science (1st ed.). Thammasat Press.

34
 35

Chapter 2
Data

ABSTRACT
The initial step for a data scientist when addressing a business question is to identify the data type,
as not all types can be employed in data mining analyses. Accordingly, the data scientist must select a
suitable data type that corresponds to the data mining technique and classify the data into categorical
and continuous types, regardless of the source of the data. Quality control is a significant factor for the
data scientist, particularly if data collection was poorly administered or designed, leading to issues like
missing values. Once the data scientist has acquired a relevant dataset, they should inspect the outliers
associated with each feature to make sure the data is suitable for analysis. Observing outliers through
data visualizations, such as scatter plots, is a common practice among data scientists, highlighting the
crucial role of data type determination.

INTRODUCTION

Though the Administrators or Domain Experts in each industry see how necessary to use data analysis to
solve problems or to drive the organization, though they and are aware of the advances in data analysis
using data mining techniques, they cannot link the existing data to individual data processing techniques.
In many cases, it was found that the development of data visualization is sufficient for data analysis to
answer questions without the need for data mining techniques. On contrary in some cases, the organiza-
tions do not have enough data to be analyzed to respond to the determined questions.
The main reason is that administrators or domain experts in the organization lack understanding
of which data can be used and how the questions should be asked in order to gain useful data. Many
organizations have developed their domain experts to become data scientists and trained them in a wide
range of data analysis. The positions of data scientists are necessary as they can use the existing data to
analyze and solve problems in a timely manner. For example, in the medical industry, patient data can
be used to analyze the treatment. The use of data in each form of data analysis is therefore an important
point for data science students.

DOI: 10.4018/978-1-6684-4730-7.ch002

Copyright © 2023, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
 
Data

ATTRIBUTE

The attributes of the data are used to define the scope of the Data Object. For instance, the data attri-
bute is gender, and data object is male, female, or genderless. Data attributes are also used in different
contexts (Inmon and Lindstedt, 2015). When data scientists write programs, data attributes are viewed
as variables and considered as features. Data scientist collects data in a dataset consisting of a feature
that stores multiple records. In data science, these records are called “Instance”. The structure to store
Feature, Data Objects, and Instances, is as follows (Gru¨tter, 2019: Angiulli & Fassetti, 2021).

Table 1. The example of feature, data object, and instance

First Name Last Name

Jirapon Sunkpho
Sarawut Ramjan
Kom Campiranon

In Table 1, there are totally 2 features consisting of first name and last name. The first name features
are Jirapon, Sarawut and Com while the last name features are Sunkpho, Ramjan and Campiranon,
respectively. When data objects from multiple features are combined, they become the Instances. From
Table 1, there are 3 Instances consisting of Jirapon Sunkpho, Sarawut Ramjan and Kom Campiranon,
respectively.
Data Dimensionality: Is the number of features within a dataset (Li, Horiguchi and Sawaragi, 2020).
In data science, it focuses on analyzing the data in various fields without focusing on storing data from the
beginning. The number of dimensions is an issue that the data scientist must consider in order to select
only the features that can support the analysis. Therefore, the data scientists need to reduce the number
of dimensions to have the features necessary for data analysis. This allows data scientists to reduce the
time and digital resources required to process massive instances of datasets.
Resolution: Is a measure used to store Data Object in each Feature. In Data Science (Wang et al.,
2019), when data scientists collect datasets from multiple sources, there is often a problem with stor-
ing the same data across multiple measures. For example, Feature height from dataset A is collected in
centimeters, whereas Feature height from dataset B is stored in inches. Therefore, before implementing
those datasets, data scientists must perform Attribute Transformation or changing the shape of the Data
Object to be in one form or another

DATA TYPE

Different data mining and visualization techniques use different types of data. It may use only one type
of data or a combination of different types. It is a great challenge for Data Scientists to analyze the data
to identify whether it is consistent with the data mining techniques and to answer the determined ques-
tions. The data can be divided into 2 types.

36

Data

Categorial Data

Categorial Data is the counted data considered as quantitative data without continuity. It also includes
Nominal Scale and Ordinal Scale. Each category has different uses (Van et al., 2005).

1. Nominal Scale: Is the data collected as grouping (Andrews and Messenger, 1973), such as gender,
including female, male, and other genders. When data scientists use nominal scale in data analysis,
they can compare data of the nominal scale; males and females as presented in the example.

Figure 1. Example of data visualization with nominal scale

As seen in the figure, data scientists can compare the Nominal Scales to describe the data with Mode
or the most duplicated data. For example, the data ranged 0-5, mode 3 means the number 3 is most cho-
sen. The data ranged 6-10, mode 7 means that the number 7 is most chosen. Data scientists should not
use nominal scale to calculate by adding, subtracting, multiplying, dividing or averaging. For example,
if the data show that there are 7.5 males, it does not make sense in data analysis because in reality there
cannot be 7.5 people. Moreover, turning letters into numbers cannot give those numbers real values to
be calculated. If Bang Rak District is valued as No. 1 and Sathorn District as No. 2, then 1 plus 2 equals
3 means nothing in data interpretation. In many cases, the nominal scale data can be used with other
data types to process visualization data as in the example.
As obvious from the figure, the Nominal Scale is the survival rate and the mortality rate of smokers
developed into visualization in the form of histogram which shows the range of the living and the dead
in the x-axis, and the frequency of the living and the dead in the y-axis.

2. The Ordinal Scale represents abstractions with ordinal numbers (Yager, 2020). Intention is an indi-
vidual emotion. Data scientists can represent that emotion as Likert Scale ranging from 1 to 5. For
example, “no intention to buy” is represented by number 1 while “probably buying” is represented
by number 2 with the intensity of buying intent being from weak to intense at number 5. When

37
 
Data

emotion is represented by numbers, data scientists can therefore draw conclusions. If the respon-
dents choose number 4 the most, this emotion cannot be determined in average value. Number 4.5
cannot be interpreted as what the level of purchase intention is. However, data scientists can apply
ordinal scale called Supervised-Learning such as Regression in data analysis, by responding to the
determined problems. For example, the purchase intention on product A has a significant influence
on the purchase intention on product B.

Figure 2. An example of the application of the nominal scale with other data

Continuous Data

Continuous data is the data presented with real numbers (Mesner and Shalizi, 2021), such as financial
data and temperature data. These data can be used for calculating the average such as the data on average
salary for employees who have been in the company for more than 5 years. Data scientists can develop
visualizations from continuous data as follows.
According to the figure, data scientists create a scatter plot from continuous data where the data in
the x-axis represents the temperature, and the y-axis represents the amount of ozone. The data of both
features are real numbers. It does not matter whether the original data that the data scientist imports
from the sources or outside are Categorical Data or Continuous Data, data scientist should consider the
objectives of the data analysis. Then, data scientists can select the suitable mining techniques for analyz-
ing the data. The desired data can then be selected.

38

Data

Figure 3. Example of visualization from continuous data

DEFINITION OF DATA TYPES WITHIN THE RAPIDMINER SOFTWARE

This book applies RapidMiner software as a tool to experiment data analysis using different data mining
techniques. According to the academic classification, data is divided into Categorial Data and Continu-
ous Data. But RapidMiner provides a method for defining each data type as follows.
In the example, the data types within RapidMiner have been divided as the following:
Polynominal: Is the data that can be in both alphabets and numbers. Data scientists can import
Categorial Data and define it as Polynominal, such as gender. However, data scientists need to consider
the data through each type of data mining techniques. Alphabetic and integer data cannot always be
processed with the same data mining techniques, although they can also be designated as Polynominals.
Binominal Data: Is a type of data that have only two values; “No” can be replaced with number 0,
and “Yes” with the number 1.

39
 
Data

Figure 4. Examples of data types within RapidMiner

Real Data: Is a real number such as temperature, height, or distance. However, there are certain types
of data analysis techniques that can analyze both Continuous and Categorial Data such as Linear Regres-
sion. The examples of such analysis are the Ordinal Scale analysis to examine the correlation between
different variables, and the real data analysis to forecast the results of data having 2 or more variables.
Integer: Is the whole number. Data scientists should be careful not to convert letters into integer
number in the quantitative data analysis.
Date_time: Is a period of time; day, month, year, and duration.
Date: Means date, day, month and year.
Time: Means a period of time.
All of Date_time: Date, and Time data are used to analyze time-controlled data such as Time Series
technique.
Therefore, accurate determination of such data types will affect the selection of appropriate techniques.
It also affects the development of data visualization as well.

DATA QUALITY

To process data with data mining techniques, the data set must be qualified for the analysis. Data mining
technique is only a processing tool, but data scientists must survey and prepare the data in consideration
of the quality of the data in many areas in order to obtain accurate results.
Bias: Is a careful selection of data analysis to reduce the high concentration of opinions among
the samples (Carlson, Livermore and Rockmore, 2020). Thailand’s public data system is operated by
state-owned energy enterprises. If data scientists would like to study the needs of citizens or the level of
satisfaction the citizen has upon the government, and they cannot collect data from people in the whole
country, the random methods must be used to collect the data. However, if the scientists only randomly

40

Data

collect data from people living in Bangkok and surrounding areas, the randomization causes biases that
the data cannot represent the whole nation.To reduce bias, the appropriate proportion of the population
must be concerned. For example, Chiang Mai Province has a denser population than Loei Province, so
data scientists cannot do random sampling with equal numbers because when the data of all provinces
are put together, it would not be sure whether such information can be used to represent Thailand. In
other cases, the proportion of the actual data is not related to the sample number. For example, data
scientists randomize only 60 percent of samples to analyze data on debt defaults; however, the real data
shows that there is only 40 percent. It means that from randomization, data scientists receive 20 percent
more than the real data.
Outlier: Refers to an abnormal Data Object (Li et al., 2019: Jyothi et al., 2020). When a data object
has an outlier, either negatively or positively, the feature average is distorted. Data scientists can use sur-
vey processes to visualize data patterns in order to detect outliers by creating a Scatter Plot visualization.

Figure 5. Exploring data objects that are outlier

As seen in the figure, Scatter Plot takes the Data Object from the Feature (Shao, et al., 2017). Total
Population which is presented on the x-axis, and the features showing the number of murders is shown
on the y-axis. From the visual data, data scientists should firstly consider whether the value is really
outlier or no, by consulting with data experts in order to verify whether the value is abnormal or differ-
ent from most data. This is to avoid a data entry error.
Missing Value: Is an empty Data Object within a Feature (Montesdeoca et al., 2019). It can occur
for many reasons. When the users do not fill out information either on paper or digital forms, it is pos-
sible that the data collection tools asking for what the users do not have data. For example, if the person

41
 
Data

filling out the information is a male but the tools require him to inform the frequency of menstruation,
then, the user may skip filling those questions. This causes missing value. In another example, if the
applicant is a student with no income, but the salary information must be filled, the user will skip filling
out the information as well.
Once a Missing Value has occurred, the Data Scientist can do 3 approaches as follows:

1. When the missing value is between 0 and 14%, data scientists can remove the instance.
2. In the case that the missing value is between 15 and 40%, data scientists can add the possible data
to replace the missing values. If the feature is continuous, data scientists can replenish the mean
value obtained from calculating all data objects to replace the missing value. Data scientists should
be concerned of the type of data before calculating the replenished value to replace the Missing
Value. The replenished value cannot be the average.
3. If the missing value is 41% or more, data scientists should consider refraining the use of feature in
data analysis because the data is not large enough to represent any data in the analysis.

Duplicate Data: Is data redundancy (Song, Yu, and Bao, 2019). In data science, this can happen
when combining more than two data sets. For example, it occurs when there is more than one instance to
indicate the same data while the only difference is the data owner’s email address. When data scientists
combine the same data collected from two data sets, there is a duplication of data. In such a case, data
scientists take data into the preparation process in order to reduce data redundancy.
Inconsistency Data: Is inaccurate data (Johnny and Trovati, 2020). Such cases can occur, when the
users fill out the wrong information in the first place; such as the use of negative numbers for human
height or a typo in the name. This can be prevented by designing an online form that alerts the user to
the misinformation before the user can enter it.

CASE STUDY

Among the vast amount of data, a number of people including beginner data scientists are confused about
what kind of data can be used for analysis, what type of problems requires only data summaries, what
type of problems requires only Classic Statistic, what type of problems requires data mining techniques to
process. Therefore, This chapter presents a collection of case studies on the use of data in various aspects.

Becoming a Data-Driven Decision-Making Organization: Seven

Challenges Facing Not-for-Profits (Hume & West, 2020)

Among the trends in Digital Transformation, organizations are aware of the use of data in decision-
making in both the public and private sectors. However, there are some articles that address 7 challenges
of non-profit organizations in using data to make decisions.

1. Selecting the Right Data for Analysis: Amidst the overwhelming data, organizations need to
first review what is the purpose of corporate data analysis. The organization can, then, collect and
store data in order to identify methods for collecting or acquiring the data. If the data cannot be
obtained from within the organization, it is necessary to determine the external sources. However,

42

Data

if an organization realizes that acquiring or storing data is not worth the investment, it can review
whether there are other sources of data to be analyzed.
2. Selecting the Right Analytical Tool: There are a number of software to be used to analyze the data.
Many organizations start their data analysis with popular spread sheet software such as Microsoft
Excel that can take datasets for compiling, calculating, and developing data. It is considered a stan-
dard tool that most employees have been trained since high school. Thus, organizations can save
a lot of training costs on such software. For organizations that expect more complex data analysis
results, the use of database data for data analysis and create more professional visual data, they can
use the software like Business Intelligence which can take data from the database to be analyzed,
preparing financial reports from the Accounting Database, and analyzing the data source before
processing it into the software such as Microsoft Power BI or Tableau.

BI is not only used for analyzing historical data but also for the future decision-making, such as prod-
uct price setting, which is calculated from the average product price in the market. To process in such a
manner, the scientists can use competitor price data to analyze in the form of time series to show price
trends and market segmentation where competitors have placed their brand positioning. However, the
afore mentioned software is still difficult to access, especially for the employees who have no experience
in data science. So, it’s important that organizations take action to explore the tools in the market and
then study which tools are suitable for your organization’s work and worth the investment. In addition,
when selecting the tools, organizations must also plan for the installation and maintenance cost of those
tools in terms of procuring hardware and the amount of software sufficient for the number of personnel.

3. Team Building for Data Analysis: While the businesses offer a high rate of paid for data scientists,
the non-profit organizations have a limited budget. So, training and appointing a Domain Expert in
a process with similar capabilities as a data scientist becomes an attractive and more budget viable
option.

However, when recruiting people, non-profit organizations can select people with potential to be
improved, such as those who graduated in information technology or finance, and those who have fun-
damental knowledge sufficient for building up to become a data scientist in the future.

4. Centralization of Data: In the real situation, organizations, especially those in smaller, less tech-
savvy organizations, tend to store data that is random and scattered. Consequently, data integration
has become a huge challenge for organizations, especially in the organizations that lack experi-
ence in structured data integration. This data integration becomes another factor for selecting both
hardware and software for future data analysis.
5. Making Accurate Data: Although data scientists have many technical methods to analyze data,
they will never be able to process data properly if the original data is not correct. Before bring-
ing data into the Database or Data File, the accuracy of the data is to be concerned first. Many
organizations often try to pass this responsibility on to their IT staff. But the person who should
understands the data accuracy is the one who is directly responsible for bringing the data into the
system, such as the financial department. Therefore, the organization needs to provide a Check
List of the validation of data as a tool for personnel to verify whether the data entered the system
is correct or not. For example, email addresses should be separated by @ sign.

43
 
Data

6. Creating a Culture of Change: To create a culture of change to drive the use of data as a basis
for decision-making, the board of directors needs key policymakers to drive it. This may start from
allocating a budget to invest in Digital Infrastructure by taking into account the balance between
hardware and software investment along with recruiting or developing personnel to use those
technologies. Organizations may start with small projects involving data analytics to pilot other
agencies. Therefore, the most important driver in creating a culture of change is a group of people
who lead an organization.
7. Initiating an Informed Decision: The use of data for decision-making does not focus on visualiza-
tion. But it is the use of the results obtained from visualization to support or drive organizational
decision-making, such as creating a Trend Chart comparing Return of Investment from Classic
marketing versus Digital Marketing from 2009 to 2022 to make decisions on whether the organiza-
tion should invest more on human resources and budget into marketing.

From the seven challenges, non-profit organizations realized that even if they do not require busi-
ness or market share expansions like any other normal business organizations, they cannot neglect the
use of data in decision-making, starting with investing in enhancing their computing capabilities with
sufficient Digital Infrastructure. It is to increase the employee skills to support data processing. It also
benefits from driving the organization with the use of data, analysis of effective strategies to drive the
organization towards its vision.

Market Analysis From Banking Customer Data (Torrens & Tabakovic, 2022)

Customer financial transactions through Banking Apps provide a valuable dataset for analyzing market
data. The bank can analyze customer patterns to develop effective marketing strategies. Machine learn-
ing itself is a modern tool to learn customer patterns. Therefore, the bank has applied machine learning
to analyze market data to develop the bank to become a Data Driven Organization.
Therefore, the Bank should aims to collect data on customers’ financial transactions, which reflect
their behavior in using money through online channels. The customer behavior on websites or social
networks show different insight data from their behavior in online shopping through the Bank’s Apps.
When the customers like to consume something, they often subscribe to social network channels such as
YouTube Channel or Instagram, but the bank’s data only show what the customers pay on a daily basis,
such as shopping for everyday things at the superstore. Incidentally, advertising within Digital Banking
Apps may not display billboards like other online services, but it has to focus on accumulating points to
use in other services, such as accumulating credit card points.
The variables used in the data analysis, therefore, cover the nature of the customer’s spending. It can
tell higher amount of money spent to buy a particular product than others, the use of money to buy the
same product from multiple sellers or competitors, and where customers conduct financial transactions
through Digital Banking Apps. Based on the variables reflecting the frequency of using this money,
banks can divide customers into high and low purchasing power groups as well as being able to distin-
guish which customers do not have behavior in conducting transactions through Digital Banking Apps.
The dataset can be used for analysis with K-Mean Clustering in order to use the collected data to
determine an objective marketing strategy, including increasing brand loyalty, increasing the digital
spending level of customers of low purchasing power customers, increasing the spending frequency of
current customers, and retaining existing customers.

44

Data

Banks can use the Supervised-Learning techniques, Linear Regression Algorithm, Logistic Regres-
sion Algorithm, or Random Forest, to predict the likelihood of customer spending. In this case, the bank
used a dataset from 150,000 customers transacting over a period of 12 months to train the model and
then predict the possibilities with various data mining techniques until it is found that the most efficient
technique is Random Forest. The bank can also group its data analysis according to the type of industry
in which customers purchase goods through bank financial transactions. In this case, the bank analyzes
and segments them into Gastronomy, shoes, sports, cosmetics, and tourism and also found that Random
Forest is the most effective method of forecasting the use of money to buy products from various industries.
Thus, this is another case study where banks can use existing datasets to analyze data to develop
marketing strategies. Other businesses can also group data with interesting variables for data analysis too.

Analyzing Data from Big Data Generated by Production Systems

to Support Decision-Making (Rogers & Kalinova, 2021)

In the manufacturing industry, there are many processes to increase efficiency, such as controlling the
production standard, reducing losses from manufacturing errors, shortening the production process,
and minimized human-controlled production. Currently, the Internet of Things Technology (IoT) is
applied to detect actions occurring in each production process, detecting the amount of liquid or gas in
a pipeline. And from the application of such technology, the industry has received real time data and it
creates a data set as Big Data. From such Big Data, the manufacturing industry can analyze the data in
a variety of perspectives, which include forecasting whether each machine has reached its maintenance
interval, improving the productivity of marketing audience analysis production, streamlining business
processes to retain customer groups, lowering the cost of raw materials in production, and optimizing
security during the production process.
To answer the above question, the manufacturing industry also has to do digital planning to support
future data analysis. It includes the adoption of mobile devices such as tablets or mobile devices to dis-
play real-time data analysis, provision of software used for data analysis such as inventory management,
movement of raw materials in production, preparation of a semi-automated production process using
robots, preparation of sensor technology to collect data in real-time, the use of data systems to control
production, and the use of models to forecast data and plan for future technological changes that can be
used in conjunction with manufacturing systems such as AI or Blockchain.
By implementing a large number of data processing technologies, the manufacturing industry is
able to reduce the number of human-machine control. It can also control the production process from
a distance via a remote system. Companies can collect data from those devices and analyze them to
solve problems in the manufacturing process, such as discovering defective equipment or loss analysis
from the production process. Mobile analytics data visualization can be done in a dashboard style. Us-
ing a dashboard, data users can compare data collected through IoT with Key Performance Indicators
to reflect process efficiency.
With this sensor technology, companies can also manage their inventory by checking whether the
amount of raw materials left in the production or the product is at the level that needs to be ordered more
or produced in order to control storage costs. Such data analysis can also be performed throughout the
Digital Supply Chain so that the end customer needs are delivered to the raw material suppliers.
The manufacturing sector also faces the challenge of equipping staff with expertise in IoT technology
and data analytics skills, as well as developing senior management understanding of how to ask ques-

45
 
Data

tions suitable for further analysis. Above all, the board of directors of each company need to formulate
a digital strategy to drive the organization towards a Data Driven Organization.

The Challenges of Being a Data Driven Organization (Bean, 2022)

Organizations worldwide are aware of being a Data Driven Organization. Those companies have adapted
hard by adopting new technology, improving processes within the delivery of products and services
through reliance on technology, and even trying to change the corporate culture until being able to compete
over competitors. The biggest challenge, however, is that large companies or even government agencies,
which are long established, have a large number of senior employees having been working within the
company. Those employees not only tolerate changes in technology or new ways of working, but also
have political power within the company to resist any form of change. The emergence of the COVID-19
crisis has caused the companies around the world to change he way they work. Senior employees are in
a state of compromise, relying on technology and digitization to proactively change the way they work.
On the other hand, consumers have various options to receive information. Therefore, consumers can
also consider choosing the best quality information to support their decision making. With data collec-
tion technologies such as IoT and high-performance data processing technologies, today companies can
bring large data sets to be processed. No more data sampling techniques are needed because today’s
digital resources are powerful enough to analyze data.
Among organizations that are trying to adapt to become a Data Driven Organization, the success factor
of the organization turns is none other than the inspiration of leaders to be a Data Driven Organization
instead of technological change. The main obstacle to organizational change is the resistance, which
arises from the organizational culture. Leaders can often start by cultivating people in the organization
to realize the value of data and initiate the management of corporate assets. To ensure that organizations
have policies and measures to protect personal data has become the global pressure from government
laws, increasingly vigilant about securing personal data to their citizens.
The solution for organizations that are trying to become a Data Driven Organization is that leaders
must have an innovative character. Among many models of data mining techniques, leaders need to create
a question-frame where data analysis initiatives differ from competitors in order to obtain answers that
competitors cannot access, and dare try and make mistake in order to create an experience of learning
from failures.

Relevance of Big Data, AI, and Robotics (Curry et al., 2022)

Europe has initiated the establishment of associations related to Big Data, AI and Robotics to strengthen
Europe’s international competitiveness. The association has faced a variety of challenges, prompting
the association itself to analyze the weakest elements in ecosystem for the promotion of Big Data, AI
and Robotics until these elements were met: Unpredictable European Policies and Measures Society’s
confidence in Big Data, AI and Robotics, knowledge and skills in Big Data, AI and Robotics, digital
utilities that still need to be developed, technological complexity and investment with a very high budget
in digital. These elements have promoted Big Data, AI and Robotics in Europe, as well as differentiating
research studies in each field and not leading to a better understanding of how to apply Big Data, AI and
Robotics in both business and technology.

46

Data

The Big Data, AI and Robotics Association has become an intermediary to connect those elements,
advocate research policy and share knowledge to connect stakeholders in terms of research and technology
implementation in more complicated environment. Therefore, the Association has developed a frame-
work for creating an ecosystem that facilitates such action, with the following components: Principles
of Rights and Values of Big Data, AI and Robotics in Europe, the Value of Big Data, AI and Robotics
to Business, Society and Citizens Policies, Regulations, Assurances and Standards for Big Data, AI and
Robotics. These principles create an innovative ecosystem used to promote the creation of knowledge
and skills in Big Data, AI and Robotics, and the ecosystem also promotes research and application of
Big Data, AI and Robotics.
By building an ecosystem of Big Data, AI and Robotics and innovations, Big Data, AI and Robotics
will promote the growth of Europe’s technology for decision support and help Europe lead the way in
science and innovation related to AI, Data and Robotics and its application, as well as to build confidence
in European citizens to implement such technology.

CONCLUSION

The dataset is made up of a combination of features which store data objects internally. And each dataset
can contain multiple instances. Using such data to analyze with data mining techniques, data scientists
need to first survey the data in order to identify the data type first. Each data mining technique requires
the use of different types of data. Additionally, in response to some types of problems, only data visu-
alization can be used without analyzing the data with data mining techniques. However, even in data
visualization development, data scientists also need to understand the data types in order to accurately
display the data and answer the determined questions.

EXERCISE QUESTIONS

1. Explain and give examples of Category Data.

2. Explain and give examples of Continuous Data.
3. In RapidMiner software, what types of data can the continuous data be defined as?
4. Explain about the bias data and give examples.
5. Describe data objects that are outliers.

REFERENCES

Andrews, F. M., & Messenger, R. C. (1973). Multivariate nominal scale analysis; a report on a new
analysis technique and a computer program. University of Michigan.
Angiulli, F., & Fassetti, F. (2021). Uncertain distance-based outlier detection with arbitrarily shaped
data objects. Journal of Intelligent Information Systems, 57(1), 1–24. doi:10.100710844-020-00624-7

47
 
Data

Bean, R. (2022). Why Becoming a Data-Driven Organization Is So Hard. Harvard Business Review
Digital Articles, 1–6.
Carlson, K., Livermore, M. A., & Rockmore, D. N. (2020). The Problem of Data Bias in the Pool of Pub-
lished U.S. Appellate Court Opinions. Journal of Empirical Legal Studies, 17(2), 224–261. doi:10.1111/
jels.12253
Curry, E., Heintz, F., Irgens, M., Smeulders, A. W. M., & Strmigioli, S. (2022). Partnership on AI, Data,
and Robotics. Communications of the ACM, 65(4), 54–55. doi:10.1145/3513000
Gru¨tter, R. (2019). A framework for assisted proximity analysis in feature data. Journal of Geographical
Systems, 21(3), 367–394. doi:10.100710109-019-00304-3
Hume, E., & West, A. (2020). Becoming a Data-Driven Decision Making Organization: Seven Chal-
lenges Facing Not-for-Profits. The CPA Journal, 90(4), 32.
Inmon, W. H., & Lindstedt, D. (2015). Data architecture: a primer for the data scientist : big data, data
warehouse and data vault. Morgan Kaufmann/Elsevier.
Johnny, O., & Trovati, M. (2020). Big data inconsistencies and incompleteness: a literature review. In-
ternational Journal of Grid and Utility Computing, 5, 705. 10.1504/IJGUC.2020.110057
Jyothi, P. N., Lakshmi, D. R., & Rao, K. V. S. N. R. (2020). A Supervised Approach for Detection
of Outliers in Healthcare Claims Data. Journal of Engineering Science & Technology Review, 13(1),
204–213. doi:10.25103/jestr.131.25
Li, J., Horiguchi, Y., & Sawaragi, T. (2020). Data Dimensionality Reduction by Introducing Structural
Equation Modeling to Machine Learning Problems. 2020 59th Annual Conference of the Society of In-
strument and Control Engineers of Japan (SICE), Instrument and Control Engineers of Japan (SICE),
2020 59th Annual Conference of the Society Of, 826–831.
Li, K., Li, J., Liu, S., Li, Z., Bo, J., & Liu, B. (2019). GA-iForest: An Efficient Isolated Forest Framework
Based on Genetic Algorithm for Numerical Data Outlier Detection. Transactions of Nanjing University
of Aeronautics & Astronautics, 36(6), 1026–1038.
Mesner, O. C., & Shalizi, C. R. (2021). Conditional Mutual Information Estimation for Mixed, Dis-
crete and Continuous Data. IEEE Transactions on Information Theory, 67(1), 464–484. doi:10.1109/
TIT.2020.3024886
Montesdeoca, B., Luengo, J., Maillo, J., García-Gil, D., García, S., & Herrera, F. (2019). A first approach
on big data missing values imputation. IoTBDS 2019 - Proceedings of the 4th International Conference
on Internet of Things, Big Data and Security, 315–323. 10.5220/0007738403150323
Rogers, S., & Kalinova, E. (2021). Big Data-driven Decision-Making Processes, Real-Time Advanced
Analytics, and Cyber-Physical Production Networks in Industry 4.0-based Manufacturing Systems.
Economics, Management, and Financial Markets, 16(4), 84. doi:10.22381/emfm16420216
Shao, L., Mahajan, A., Schreck, T., & Lehmann, D. J. (2017). Interactive Regression Lens for Exploring
Scatter Plots. Computer Graphics Forum, 36(3), 157–166. doi:10.1111/cgf.13176

48

Data

Song, J., Yu, Q., & Bao, R. (2019). The Detection Algorithms for Similar Duplicate Data. 2019 6th
International Conference on Systems and Informatics (ICSAI), Systems and Informatics (ICSAI), 2019
6th International Conference On, 1534–1542.
Torrens, M., & Tabakovic, A. (2022). A Banking Platform to Leverage Data Driven Marketing with
Machine Learning. Entropy (Basel, Switzerland), 24(3), 347. doi:10.3390/e24030347 PMID:35327858
Van der Ark, L. A., Croon, M. A., & Sijtsma, K. (2005). New developments in categorial data analysis
for the social and behavioral sciences. Lawrence Erlbaum Associates Publishers.
Wang, J., Sato, K., Guo, S., Chen, W., & Wu, J. (2019). Big Data Processing With Minimal Delay and
Guaranteed Data Resolution in Disaster Areas. IEEE Transactions on Vehicular Technology, 68(4),
3833–3842. doi:10.1109/TVT.2018.2889094
Yager, R. R. (2020). Ordinal scale based uncertainty models for AI. Information Fusion, 64, 92–98.
doi:10.1016/j.inffus.2020.06.010

49
50

Chapter 3
Software Installation and
Introduction to RapidMiner

ABSTRACT
Licensed and open-source data mining software has made it easier for data scientists to process and
analyze large datasets. Python and R are popular open-source tools used for teaching data mining
techniques, including data classification and clustering. However, one of the main challenges for
domain experts looking to upskill in data science is computer programming. Licensed software such
as RapidMiner provides a solution to this issue, as it is designed for data mining processing without
requiring computer programming skills. RapidMiner allows data scientists to bring their data sets and
mining models into an analysis process that includes all data science mechanisms. It also offers vari-
ous supporting models for data pre-processing and creates data visualizations during data exploration.
After data mining, RapidMiner displays easy-to-interpret results, enabling data scientists to effectively
explain the answer to a business question to stakeholders. With its user-friendly interface, RapidMiner
is an excellent choice for those without a technical background in programming.

INTRODUCTION

RapidMiner software is very popular in data science work, and has been used in data scientist devel-
opment trainings (Chisholm, 2013: Klinkenberg & Hofmann, 2014: Kotu & Deshpande, 2014). The
scientists can learn about various data mining techniques and apply those concepts in data analysis
with RapidMiner without writing any program. Moreover, RapidMiner software supports every step of
the data science field work. Data scientists can import datasets into RapidMiner software to promptly
explore data through the feature visualizations. In data preparation, the scientists can use models within
the RapidMiner software to customize their data ready for analysis with various data mining techniques.
RapidMiner software provides the sample data sets for the scientists to use in experimenting data during
their trainings as well.

DOI: 10.4018/978-1-6684-4730-7.ch003

Copyright © 2023, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Software Installation and Introduction to RapidMiner

RAPIDMINER SOFTWARE INSTALLATION

Professors and learners of data mining techniques who would like to use RapidMiner software as a tool
to study and experiment for data analysis can download RapidMiner software at the following this link,
https://rapidminer.com/, which will take you to the download page as seen below (Chavan, 2020: Soft,
2020: Vasudevan, 2021).

Figure 1. RapidMiner software download

As seen in the figure, select Apply Now at RapidMiner in the form of Education Program, then fill
in the details to verify your identity, as seen below.
From figure 3.2, select Educational Purpose, and then fill in your first and last names, educational
institution name, the role of the software user consisting of teachers and students, email address and
password (Naparat, 2020). Then press the button Register.
As seen, my.RapidMiner.com will send an e-mail requiring the user to confirm the software download
request. The user, then, should verify the e-mail.
From the picture, once the email is verified, my.RapidMiner.com will download the software as
shown in the picture down here.
From the figure, check the type of operating system inside the computer. In the case of Windows
operating system where there are 2 types; 32 Bit and 64 Bit, select the correct download according to
the operating system. And in the case of MAC OS or Linux operating system, choose to download ac-
cording to the operating system. Once the download is completed, you will receive a file for the software
installation.
When double clicking on the Installation File, the start screen will appear. Then press the Next >
button to go to the next step.

51
 
Software Installation and Introduction to RapidMiner

Figure 2. Filling out the information to verify your identity

Figure 3. Sending an e-mail for confirmation

From the figure, read the License Agreement and press I Agree to go to the next step.
At this step, select the file location to install the software. Then press the Install button.
After completing the software installation process, press Finish to start using RapidMiner software.

52

Software Installation and Introduction to RapidMiner

Figure 4. Confirmation (verify) of software download via electronic mail

Figure 5. RapidMiner software download

53
 
Software Installation and Introduction to RapidMiner

Figure 6. RapidMiner software installation

Figure 7. Starting software installation

54

Software Installation and Introduction to RapidMiner

Figure 8. License agreement

GETTING START WITH RAPIDMINER SOFTWARE

Data scientists can begin their RapidMiner software usage by logging into RapidMiner software. A
window will appear; then, the data scientists can begin modeling by selecting Blank Process as follows.
When entering RapidMiner software, there are 5 components; 1. Main Menu 2. View 3. Repository
4. Operator and 5. Parameter. Each component supports the users as seen in the image below.
The users can create files, which stores the models within the RapidMiner software called Process.
It is to access each file stored within the computer, record the process, and process the data mining from
the menu and the icon in the Main Menu section.
Data scientists can create data mining models in the Design section. When processing the model, the
results of the data analysis can then be checked in the Results section.
Data scientists can import the datasets to analyze into RapidMiner software by importing the data. The
scientists can access the recorded processes and retrieve the ready-made datasets provided by RapidMiner
software, so that data scientists can create models from the dataset (RapidMiner, Inc., 2020).
Data scientists can choose different types of data mining techniques from Operators. These techniques
are diagrammatic with arrangement and connection to each part according to the designed processing
algorithm.

55
 
Software Installation and Introduction to RapidMiner

Figure 9. Selecting file location

To take each piece of the data mining models to arrange and connect, data scientists can customize
the default parameters in order to control processing operations for more accurate data analysis. Data
scientists can practice using RapidMiner, starting with a sample dataset called Iris to create a process:
In the figure, data scientists run through the repository by accessing the folder where the sample
dataset, named Iris, is stored, and then import the Iris dataset into modeling. The data scientists then
draw a line to connect the output from the Iris dataset to the Result. When executing, you will receive
the result of processing as shown in the image below.
From the data set, data scientists can start exploring the data by examining the patterns of the data
in the Statistics section:
From the figure, data scientists can begin exploring the data via statistics, which provides a detailed
breakdown of the data in every feature. From the sample, Iris dataset consists of 7 features, with the
target variables holding information of the Irish flower species, as well as the other features. The sizes
of the Irish flowers are listed in statistics as it also shows the smallest, the largest value and the average
values. And from the initial examination of the data, data scientists are able to explore the nature of
data with greater details through the development of visualization (Ironfrown, 2016: Global Mindset
University, 2020: Tuition, 2021).

56

Software Installation and Introduction to RapidMiner

Figure 10. The installation is completed

Figure 11. Getting started with RapidMiner software

57
 
Software Installation and Introduction to RapidMiner

Figure 12. Main menu

Figure 13. View

From the figure, data scientists developed a box plot visualization by selecting to display the data
from features named a1, a2, a3, and a4, respectively. From the visual data, the scientists receive the in-
formation about the least value being 25%, the moderate value being 75%, and the largest value for each
feature. Data scientists can also customize the visualization display format for exploring the data quickly.

58

Software Installation and Introduction to RapidMiner

Figure 14. Repository

Figure 15. Operators

59
 
Software Installation and Introduction to RapidMiner

Figure 16. Parameters

Figure 17. Importing the data set into the Process

60

Software Installation and Introduction to RapidMiner

Figure 18. Results of the Iris dataset

Figure 19. Exploring data models with statistics

61
 
Software Installation and Introduction to RapidMiner

Figure 20. Development of box plot visualization data

CASE STUDY

The Selection of Software for Data Science Processing

(Abbasnasab Sardareh et al., 2021)

Although the industries worldwide are waking up to prepare for Big Data processing by recruiting
large numbers of data scientists, one challenge is choosing the right software for data processing. The
software packages have hidden complex statistical formulas within, whereas data scientists can simply
mouse-click on a tool to allow the software to perform the required processing. However, industries and
data scientists specializing in statistical processing still need someone who has expertise in each field
to be able to ask sharp questions and choose the right processing method to answer them. Therefore,
the industries are facing problems in transforming Domain experts, who lack experience in statistical
computing, into data scientists. For instance, domain experts in social sciences such as education, sustain-
ability are not skilled at calculating with statistical formulas (But, of course, they specialize in qualitative
data analysis and not always at quantitative data analysis. Some are very proficient.) Normally, those
who are beginners in statistics need to know about descriptive analysis, data visualization, hypothesis
testing by t-test, ANOVA, Regression analysis, and chi-squared. It has become an important variable
that data scientists use when selecting appropriate software. For some software brands, users can click
the mouse to instruct the software to process the data set. It is a package that the software has provided
processing feature. But for some brands, the users need to write a set of instructions to execute the data

62

Software Installation and Introduction to RapidMiner

set in exchange for processing it in such a way or called “Customization”. Therefore, one brand of data
processing software cannot meet all the needs of the users.
SPSS is a widely used statistical data processing program and is often the default software for data
scientists to study data processing. SPSS is capable of processing quantitative data into statistical results
such as Regression Analysis, and able to process data in Graph format. However, SPSS is a high-priced
licensed software. Many universities around the world, therefore, prefer to purchase them for their stu-
dents to use as a site license. SPSS is a software package that requires data scientists to prepare clean
datasets before they are importing into SPSS. The scientists can click the command buttons to allow
SPSS to process data as needed.
R programming language is a Free Software that data scientists can use for processing data sets
through writing instruction sets. Within the R language, there is a library where the data scientists can
run a ready-made instruction set to process, such as creating a Data Visualization with Customization,
or data processing with some data mining techniques such as K-NN. However, an interesting point is
that data scientists using the R language to process data are required to have programming skills in a
computer language before.
The variables data scientists take into account when selecting the software can be divided into 2
variables, consisting of the variable for data processing and for usage. Regarding processing, the data
scientists focus on computing capabilities in data science tasks, including a user-friendly user interface,
a number of statistic features covering statistical data processing, tools to create Data Visualization that
are commercially beautiful and valuable in data analysis. It also supports for Data Pre-Processing within
one software without relying on other software to pre-process the data before importing the data into the
software. In terms of usability, data scientists focus on processing speed and simple steps per processing
time, and simplicity in writing a set of instructions for processing.

CONCLUSION

Data scientists can learn how to use RapidMiner for data mining by installing the Education Program
RapidMiner software, designed for teachers and students to be able to take sample datasets and sample
models to practice analyzing data in different ways. The software also supports other steps in the data
analysis, such as creating data visualizations for exploring data in various formats. RapidMiner software
is a data analysis tool which the data scientists can use in their research by going beyond the limitations
of computer programming skills. However, RapidMiner software is a tool that supports data analysis in
the form of models, so data scientists need to study and understand how data mining techniques work in
order to be able to choose a sample data set appropriate for the processing of each data mining technique.

EXERCISE QUESTIONS

1. Install RapidMiner software on your computer.

2. Retrieve the Iris dataset and develop the visualization data in histogram and boxplot from that
dataset.
3. From number 2, describe the survey results from the visualization.

63
 
Software Installation and Introduction to RapidMiner

REFERENCES

Abbasnasab Sardareh, S., Brown, G. T. L., & Denny, P. (2021). Comparing Four Contemporary Statistical
Software Tools for Introductory Data Science and Statistics in the Social Sciences. Teaching Statistics:
An International Journal for Teachers, 43.
Chavan, A. (2020). RapidMiner Installation on Windows. Retrieved 15 December 2021 from https://
www.youtube.com/watch?v=4RpepmDH-7s
Chisholm, A. (2013). Exploring Data with RapidMiner: Explore, Understand, and Prepare Real Data
Using Rapidminer’s Practical Tips and Tricks. Packt Publishing.
Global Mindset University. (2020). RapidMiner Data Exploration demo | Data Analysis - Using the Iris
Data | Data Mining. Retrieved 15 December 2021 from https://www.youtube.com/watch?v=EfQkqcj3MIk
Ironfrown. (2016). RapidMiner Stats (Part 5): Boxplots. Retrieved 15 December 2021 from https://www.
youtube.com/watch?v=IJ180B4ZiLo
Klinkenberg, R., & Hofmann, M. (2014). RapidMiner: data mining use cases and business analytics
applications. CRC Press.
Kotu, V., & Deshpande, B. (2014). Predictive analytics and data mining: concepts and practice with
RapidMiner. Elsevier Ltd.
Naparat, D. (2020). Data Mining with Rapidminer - Getting RapidMiner Educational License. Retrieved
15 December 2021 from https://www.youtube.com/watch?v=hBxjN1YYXXU
RapidMiner, Inc. (2020). How to Import Data into a Repository | RapidMiner Studio. Retrieved 15
December 2021 from https://www.youtube.com/watch?v=DS-tYhgA5lA
Soft, N. (2020). RapidMiner Tutorial For Beginners | Download & Install | Introduction to RapidMiner.
Retrieved 15 December 2021 from https://www.youtube.com/watch?v=im11NBJfjhM
Tuition, D. (2021). RapidMiner Tutorial, Learn Data Cleaning, Data Visualization & Data Analysis
with Rapidminer. Retrieved 15 December 2021 from https://www.youtube.com/watch?v=r27eUHJtIFs
Vasudevan, S. (2021). Rapid Miner Installation - Step by Step explanation. Retrieved 15 December 2021
from https://www.youtube.com/watch?v=Wp42fV4q66Y

64
 65

Chapter 4
Data Pre-Processing and
Example of Data Classification
With RapidMiner

ABSTRACT
In this book, the focus is on data mining with RapidMiner. However, it’s important to note that there are
other essential steps to consider when delving into the realm of data mining. This chapter serves as an
introduction to the process of data pre-processing using RapidMiner, allowing readers to practice with a
data set example available on the platform. With RapidMiner, data pre-processing begins with exploring
the data visually and then selecting the features that will be analysed with each data mining technique.
Managing missing values in a feature is also a crucial step in this process, which can be achieved by
either eliminating or replacing them with appropriate values. In addition, RapidMiner allows data
scientists to detect outliers and normalize features easily using diagram design, without requiring any
computer programming skills. To help readers become familiar with the tools offered by RapidMiner, a
classification technique will be demonstrated step-by-step in the book.

INTRODUCTION

RapidMiner is a tool supporting throughout the process of data science work (Mat, Lajis & Nasir, 2018).
Data scientists can import data sets into RapidMiner software in order to prepare the data ready for pro-
cessing with various data mining techniques (Dai et al., 2016: Samsani, 2016: Phan, Wu & Phan, 2021).
In Missing Value Management and Outlier Management, when the data is available, data scientists can
use various data mining models, either Supervised Learning or Unsupervised Learning (Cai et al., 2016:
Mandhare & Idate, 2017: Susanti & Azizah, 2017: Abu-Soud, 2019). Processing algorithms are designed
through model connections, and then the result of data processing is executed. To connect such models,
the need for computer programming is unnecessary. Therefore, the scientists can quickly modify models
at each step and perform reprocessing to improve the model’s processing accuracy.

DOI: 10.4018/978-1-6684-4730-7.ch004

Copyright © 2023, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
 
Data Pre-Processing and Example of Data Classification With RapidMiner

DATA PRE-PROCESSING USING RAPIDMINER SOFTWARE

To import datasets into a data mining model with RapidMiner software, data scientists can manipulate
the data with Microsoft Excel. For example, Attribute Transformation is done to change the shape of the
data before it is imported into RapidMiner software. They can also choose the operator within Rapid-
Miner as well. An experiment can be performed to prepare the data as follows:

Figure 1. Using the Titanic sample dataset

Data scientists can choose to use the Titanic sample dataset for experimentation. After the dataset is
imported into the process, data scientists connect the dataset’s output to the Result section, and execute
the model. The result is as follows.

Figure 2. Titanic dataset details

66

Data Pre-Processing and Example of Data Classification With RapidMiner

As seen in the data set details, the dataset consists of 1,309 records, and the missing values ​​appear
within the dataset. Within RapidMiner, a question mark is used to represent the Missing Values. When data
scientists access the Age statistics section, it reveals the details of the Missing Values, as shown below.

Figure 3. The missing value details of age

When looking at Age, there are 263 Missing Values. But to analyze the data, Age is required. Data
scientists need to deal with the missing values in some ways. Data scientists can then press the Open
Visualizations button to display the visual information as seen below.
From the figure, data scientists can consider the data distribution with histogram visualization which
displays data of age ranges on the x-axis, and displays the data of the frequency on the y-axis. When
considering the Titanic dataset, data scientists focus on the characteristics of the survivors and casualties
from the Titanic shipwreck. Therefore, data scientists are necessary to consider the features useful for
data analysis. And when considering from the datasets, some features should be excluded. This is due
to 2 reasons. Firstly, the features on Passenger Names, Ticket Numbers and Ports of Embarkation which
inform about passenger’s destination should be eliminated because it does not affect the survival rate.
And the other reason is that some data, namely Cabin Numbers and Life Boat Numbers, have a large
number of missing values, which affects the accuracy of the data analysis. The data scientists can use an
operator called Select Attributes to remove unwanted features from the data set as shown in the picture.
As obviously seen in the picture, data scientists connect models in the dataset to the Select Attribute,
and then connect the Select Attribute to the Result. Meanwhile, the data scientists determine the param-
eters by selecting the Attribute Filter Type as the subset, then selecting the Select Attributes, and then
selecting only the features they want. This process is called Feature Selection.

67
 
Data Pre-Processing and Example of Data Classification With RapidMiner

Figure 4. Visualization of age data

Figure 5. Select attributes

As a result of Feature Selection, when executing a model within the dataset, only the selected fea-
tures are left. This is to prepare the data before processing with data mining techniques as shown in the
picture below.
To operate the Feature Selection using models within the RapidMiner software, data scientists can
select the features and also preserve the original dataset. However, within the datasets there can be
missing value data. Therefore, data scientists can use the Operator named Replace Missing Value to
manipulate the missing values.
Data scientists can manipulate the missing values for a specific feature by setting the Attribute Filter
Type as the subset and select the Select Attributes. From the sample, the scientists select Age, then
determine the default value as Average. Therefore, the missing value data within Age will be replaced
by the mean as shown in the figure.
When manipulating the missing value data, data scientists must consider the outlier data. As seen
in the example, the feature is called Passenger Fare or unusually expensive ticket prices which can be
considered from the visual information as follows.

68

Data Pre-Processing and Example of Data Classification With RapidMiner

Figure 6. Feature selection

From visual histogram data, data scientists can observe a range of data that do not clump with other
data, and have a low frequency. Therefore, such data may be considered as outlier, and can be manipu-
lated using a model named Detect Outlier (Distances) as follows.
In the figure, data scientists can examine the outlier data and obtain the results from execution as
seen below.
About the added feature, Outlier, data scientists can determine whether the data is an Outlier or not.
If it is, the feature shows True, and if it is not, the feature shows False.
To determine Parameter in Number of Neighbors is to compare the data with 10 other numbers. If there
is difference, it will be outlier. Number of Outliers refers to the amount of data that the data scientists
consider as Outlier. In this case, the scientists can determine 100 outliers. After validating the outlier
data, the scientists can remove the outlier from the dataset using a Filter Examples model.
Once the Filter Examples are imported into the Process, data Scientists define the conditions for
removing the Outliers from the Data Set by assigning Parameters in the Filters section, and select the
Outliers to equal False.
When the model is executed, the dataset has the outlier value as false, in order to store only the data
that are not outlier.
Data scientists can then standardize data by Normalization as seen in the example. Data scientists
need Normalization Feature called Age. The scientists can modify the age to range from 0 – 1. To nor-
malize model, it can done as follow.

69
 
Data Pre-Processing and Example of Data Classification With RapidMiner

Figure 7. Data sets from feature selection

Figure 8. Manipulating with missing value

From the figure, data scientists can choose the type of data processing by adjusting the parameters,
as presented below.
In the figure, data scientists select the Attribute Filter Type as the Subset, then press the Select At-
tribute button and the feature named Age. Then select Method to adjust the value within Age to be Rang
Transformation in order to adjust the age data to range from 0 – 1. The result of executing is shown in
the picture.
Therefore, in the process of surveying and preparing data, the scientists must explore data through
visual data first. The missing value and outlier data are then manipulated with further processing (Lu-
engo et al., 2020).

70

Data Pre-Processing and Example of Data Classification With RapidMiner

Figure 9. Determining parameter for replacing missing value

Figure 10. Titanic data set after manipulating the missing value

71
 
Data Pre-Processing and Example of Data Classification With RapidMiner

Figure 11. Exploring the outlier data

Figure 12. Outlier inspection

EXAMPLES OF DATA CLASSIFICATION WITH RAPIDMINER SOFTWARE

In data classification, data scientists pick up problems in data analysis from executives or data experts
(Başarslan & Argun, 2018: Zuo & Guo, 2019). The problems are then defined, which is to determine the
data sources and the datasets in order to find answers to the questions. A good example of classification
is a case study of classification of the Titanic wreck survivors. Data scientists can choose to use sample
datasets within the RapidMiner software, including pre-prepared datasets to teach machine learning
and the original dataset containing target variables. Using datasets for teaching and testing follows the
principles of data analysis, which is classified as Supervised Learning. The process is as seen below.
Data scientists analyze data using the Titanic Training dataset, which is a sample dataset within
RapidMiner software. It is used to testify various forms of data analysis, as seen in the example. The
scientists use the Decision Tree model, a technique for classifying data on the deaths and survivors of
the Titanic wreck (Yang, 2019). The obtained result is seen in the following models.

72

Data Pre-Processing and Example of Data Classification With RapidMiner

Figure 13. Outlier data display

Figure 14. Determining the number of outliers

Figure 15. Removing outlier data from the datasets

73
 
Data Pre-Processing and Example of Data Classification With RapidMiner

Figure 16. Defining conditions in order to remove outliers from datasets

Figure 17. Datasets where the outlier data are removed

74

Data Pre-Processing and Example of Data Classification With RapidMiner

Figure 18. Normalization

Figure 19. Assigning parameters to normalize

Based on the figure, data scientists set the parameter as 5 leaf nodes. The analysis shows that the
mortality rate is higher in males than females. Data scientists then need to test the model’s performance
using the dataset. The scientists use the Titanic dataset as an example. The original dataset tests and
determines the target variables as Survived in order to analyze the data.
From the picture, in order to test the performance of the model, the scientists determine target vari-
ables using the Set Role model, and then set Survived as a label or target variable. The data can be then
analyzed using machine learning capabilities with the Apply Model.
From the picture, data scientists can use the Decision Tree that have been taught. The taught machine
can test the Titanic dataset to identify which passengers died or survived. The results obtained from the
test are as follows.
The scientists can compare the actual data, used to teach the machine learning, to classify the data
whether each passenger survived or died. To test the classification accuracy, data scientists can use the
Performance model.
After analyzing the data, data scientists can test the classification accuracy of machine using the
Performance model and obtain the following results.

75
 
Data Pre-Processing and Example of Data Classification With RapidMiner

Figure 20. Dataset which are normalized

Figure 21. Data analysis using decision tree

In the figure, the efficiency of machine learning classification is 78.76 percent, and the error is 21.24
percent. In data analysis by classification, data scientists can use RapidMiner software to analyze data
through modeling without any need for programming (Ghous and Kovács, 2020). The scientists can use
the useful datasets for data analysis to teach machine learning. This will lead to efficient data testing.
This book introduces various data mining techniques to test, in order to obtain the results to the questions
by using RapidMiner as a data analysis tool.

76

Data Pre-Processing and Example of Data Classification With RapidMiner

Figure 22. Results from decision tree analysis

Figure 23. Target variables (label)

CASE STUDY

Automated Data Preparation (Zagatti et al., 2021)

Preparing the data before processing with data mining techniques starts with exploring whether the data
is complete and consistent with the original dataset. Later, data scientists need to understand the nature
of the data on whether it can actually be used for analysis to answer the questions. At this stage, the sci-
entists can also create Attribute Transformations to change the format of the data so that it can be used
for processing or collecting data from a variety of sources before then making Standardization to make
the same data in different storage formats. The distribution of the data is then analyzed to determine
how suitable it is to analyze. If the mean of the data is the outlier data, data scientists need to deal with

77
 
Data Pre-Processing and Example of Data Classification With RapidMiner

those outliers. If the data is missing values, data scientists need to deal with those missing values as
well. As is obvious, it requires a lot of the skill of a data scientist to pre-process the data. In addition, if
the data scientist is not a Domain Expert who specializes in the data set, then it is necessary to rely on
Domain Expert to analyze and interpret the results of data pre-processing at every step, which is even
more complex. Although today’s data pre-processing software such as Auto-Sklearn can be used for
automated data cleaning, the aforementioned instruction set still cannot deal with data in the manner of
categorical data. On the other hand, within RapidMiner provides an Auto Model to support scientists
to perform a wide range of data mining models by traversing complex data mining techniques but the
Auto Model does not cover all Data Pre-processing. Therefore, no matter what software a data scientist
uses in data processing, in Data Pre-Processing, the scientists still have to use their own abilities and
rely on Domain Experts.

Figure 24. Using the apply model

In 2021, there was a research article suggesting the use of meta-learning for the preparation of Data
Pre-processing in an automated format (Zagatti et. al., 2021) to help optimize data pre-processing so
that data scientists can encounter diverse and complex datasets.The technique uses a method of learn-
ing from the current dataset and then forecasting the data to complete the dataset. This algorithm starts
with filling the data by adding mean, median, and frequency values into the missing value data. This
research report uses One-Hot Encoding technique to turn those data into numbers. In the next step, this
algorithm rescales all data to the same standard by Standardization in order to convert all data to 0 – 1.
This research has selected several datasets suitable for testing the efficacy of the algorithm and then
measuring the efficiency of data cleaning using the Random Forest technique. Then, the new data set is
forecasted using k-NN technique.

78

Data Pre-Processing and Example of Data Classification With RapidMiner

Figure 25. Results from decision tree data analysis

Figure 26. Model performance test

It can be seen that Data Pre-Processing is a challenging process that combines both data mining tech-
niques, and the expertise of data scientists and Domain Expert. Data science requires a lot of practices
and experiences.

79
 
Data Pre-Processing and Example of Data Classification With RapidMiner

Figure 27. Model accuracy

Extracting Data from Social Media for Analysis (Zadeh et al., 2021)

Amidst the transition from data warehouse collection, data scientists are facing with Big Data collection.
More and more businesses are gearing up to be data-driven organizations and have discovered that social
media is the source of the big data. Businesses can access data sources on social media and collect them
to extract only the parts they need for data pre-processing. It is then analyzed with data mining techniques
to obtain results consistent with business problems. There are a number of tools from different vendors
that support Big Data analytics such as IBM or Cloudera.
In 2021, there was a research on extracting data from social media for data analysis (Zadeh, Zolbanin
& Sharda, 2021). The researchers aimed for creating a tool to support data analysis courses, especially
the Big Data for teachers to use as a teaching tool for students. The researchers have used a lot of tools
to create a framework to be used as a teaching tool, including 1. InfoSphere BigInsights from the IBM
which has a Big Data processing environment based on Hadoop, 2. JSON Query Language (JAQL)
to process semi-structured data, 3. BigSheets that supports Data Exploration and Data Visualization
on Big SQL, 4. Annotation Query Language (AQL) for Text Analytics to analyze semi-structured and
unstructured text data, and 5. Big R for processing data on Hadoop.
Such a research has been carried out Five datasets out of 180. All datasets were tested and 5 datasets
were selected, from the Accuracy, for experimentation. For example, the Titanic dataset or the Bike
Buyers dataset consists of 1. Missing Value Dataset and Categorial Dataset, 2. Missing Value Dataset
and non-Categorial Dataset 3. Non-Missing Value Dataset and Categorial Dataset 4. Non-Missing Value
Dataset and non-Categorial Dataset, and 5. Missing Value Dataset only for numeric features and Features
that are Categorial Data.
The study selected five datasets for the experiment because this research aims to test the efficiency of
the model in extracting data from social media during the data pre-processing process. With the use of
a variety of tools such as Auto-sklearn, the researchers compared the extraction efficiency from several
methods with Accuracy value.
From such efforts, it can be seen that the Data Pre-Processing is diverse, and in many cases data
scientists need the right tools for each dataset and each data analysis purpose. No machine perfectly fits
all, so data scientists need to study and apply them for each task.

CONCLUSION

Data science work starts with questions asked by executives from the business sectors or data experts in
other industries, such as the medical industry or agricultural industry. After receiving those problems,

80

Data Pre-Processing and Example of Data Classification With RapidMiner

the scientists seek information sources consisting of internal data or data from outside the organization.
Then, the dataset is taken to survey the data in order to determine whether the dataset can answer the
questions via the process of data pattern analysis, Missing Value analysis and Outlier analysis. Then,
the data is prepared for analyzing with data mining techniques; such as manipulating the Missing Value,
manipulating the Outlier and Data Normalization. Data techniques can be used to respond to the ques-
tions. Data scientists can also examine how accurate the model is. During the process, the data scientists
can also return to data preparation for more accurate data analysis. Once the have been obtained, the
scientists can use the results to present to the questioners. In this whole process, the scientists can use
RapidMiner software to support work from the start.

EXERCISE QUESTIONS

1. Manipulate the Missing Value using the Titanic dataset.

2. Manipulate the Outliers using the Titanic dataset.
3. Classify the data with the Decision Tree model using the Titanic Trained dataset.
4. Test the classification efficiency of the data using the Titanic dataset with the Titanic Trained
dataset.

REFERENCES

Abu-Soud, S. M. (2019). A Novel Approach for Dealing with Missing Values in Machine Learning
Datasets with Discrete Values. 2019 International Conference on Computer and Information Sciences
(ICCIS), 1-5. 10.1109/ICCISci.2019.8716430
Başarslan, M. S., & Argun, İ. D. (2018). Classification Of a bank data set on various data mining plat-
forms. 2018 Electric Electronics, Computer Science, Biomedical Engineerings’ Meeting, 2018, 1–4.
Cai, S., Sun, R., Hao, S., Li, S., & Yuan, G. (2016). An efficient outlier detection approach on weighted
data stream based on minimal rare pattern mining. China Communications, 16(10), 83-99.
Dai, H., Zhang, S., Wang, L., & Ding, Y. (2016). Research and implementation of big data preprocess-
ing system based on Hadoop. 2016 IEEE International Conference on Big Data Analysis (ICBDA), 1-5.
10.1109/ICBDA.2016.7509802
Ghous, H., & Kovács, L. (2020). Efficiency Comparison of Python and Rapidminer. Multidiszciplináris
Tudományok, 10(3), 212–220. doi:10.35925/j.multi.2020.3.26
Mandhare, H. C., & Idate, S. R. (2017). A comparative study of cluster based outlier detection, distance
based outlier detection and density based outlier detection techniques. 2017 International Conference
on Intelligent Computing and Control Systems (ICICCS), 931-935. 10.1109/ICCONS.2017.8250601
Mat, T. M., Lajis, A., & Nasir, H. (2018). Text Data Preparation in RapidMiner for Short Free Text
Answer in Assisted Assessment. 2018 IEEE 5th International Conference on Smart Instrumentation,
Measurement and Application (ICSIMA), 1-4.

81
 
Data Pre-Processing and Example of Data Classification With RapidMiner

Phan, Q.-T., Wu, Y.-K., & Phan, Q.-D. (2021). An Overview of DataPreprocessing for Short-Term Wind
Power Forecasting. 2021 7th International Conference on Applied System Innovation (ICASI), Applied
System Innovation (ICASI), 2021 7th International Conference On, 121–125.
Samsani, S. (2016). An RST based efficient preprocessing technique for handling inconsistent data. 2016
IEEE International Conference on Computational Intelligence and Computing Research (ICCIC), 1-8.
10.1109/ICCIC.2016.7919591
Susanti, S. P., & Azizah, F. N. (2017). Imputation of missing value using dynamic Bayesian network
for multivariate time series data. 2017 International Conference on Data and Software Engineering
(ICoDSE), 1-5. 10.1109/ICODSE.2017.8285864
Yang, F.-J. (2019). An Extended Idea about Decision Trees. 2019 International Conference on Com-
putational Science and Computational Intelligence (CSCI), 349-354. 10.1109/CSCI49370.2019.00068
Zadeh, A. H., Zolbanin, H. M., & Sharda, R. (2021). Incorporating Big Data Tools for Social Media
Analytics in a Business Analytics Course. Journal of Information Systems Education, 32(3), 176–198.
Zagatti, F. R., Silva, L. C., Dos Santos Silva, L. N., Sette, B. S., Caseli, H. de M., Lucredio, D., & Silva,
D. F. (2021). MetaPrep: Data preparation pipelines recommendation via meta-learning. 2021 20th IEEE
International Conference on Machine Learning and Applications (ICMLA), Machine Learning and Ap-
plications (ICMLA), 2021 20th IEEE International Conference on, ICMLA, 1197–1202.
Zuo, L., & Guo, J. (2019). Customer Classification of Discrete Data Concerning Customer Assets Based
on Data Mining. 2019 International Conference on Intelligent Transportation, Big Data & Smart City
(ICITBS). 10.1109/ICITBS.2019.00093

82
 83

Chapter 5
Classification

ABSTRACT
In the world of data mining, classification reigns supreme as a popular technique for supervised learn-
ing. Its ability to identify patterns in data by dividing it into training sets and utilizing machine learning
makes it an essential tool in answering critical questions related to data. For instance, classification can
aid businesses in identifying customers with high purchasing potential. One of the standout features of
classification is k-nearest neighbors (k-NN), which allows data to be classified according to the train-
ing data set. Decision trees are also commonly used to support decision making by producing easily
interpretable diagrams. RapidMiner is an outstanding data mining tool that can employ a range of clas-
sification techniques, including k-NN, decision trees, and naïve Bayes. In this book, readers can follow
a step-by-step guide to using these techniques with RapidMiner to achieve effective data classification.

INTRODUCTION

Data classification is the intension to classify or identify the data such as the classification of customers
who are likely to change mobile phone companies. The results obtained from the analysis are the Discre-
ate or Categorial Data, which indicates the cluster or the type of data (Vichi, Ritter & Giusti, 2013). In
data science, this group of data is referred as a Class Label. To have a variable target is to classify data
using the principle of Supervised-Learning in data processing by dividing the data into 2 parts (Mishra,
& Vats, 2021). Part 1 is for teaching machines to learn, and part 2 is to test the performance of the model.
The data classification techniques are processed as follows.
As seen in the figure, data scientists can use Classification Algorithm such as K-Nearest Neighbor
(k-NN) or Decision Tree to create Classification Model using Training Data to teach the machines to
learn in order to obtain the needed results from data classification (Liu, 2021: Mladenova, 2021). Then,
Test Data is used to apply the model to test the model’s accuracy performance. The test data and the
results obtained from the classification are compared.

DOI: 10.4018/978-1-6684-4730-7.ch005

Copyright © 2023, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
 
Classification

Figure 1. Data analysis with classification

GENERATING TRAINING AND TEST DATA SET

The main principle of Supervised-Learning is to divide the data into 2 parts, consisting of Training Data
Set and Test Data Set.

Holdout Method

To divide the data set by Holdout Method, the data is divided into 2 parts including Training Data Set
70% and Test Data Set 30%. However, if the Holdout method is used in cases where the data set is small
and is still allocated for model testing, the model lacks the opportunity to learn the nature of the data
and ultimately reduces the processing accuracy.

Cross Validation

Cross Validation method is to determine the number of rounds of division into k cycles by dividing
the data into 2 parts in every cycle (Mnich et al., 2020). For example, the number of data division is
determined and k is equal to 4. Therefore, in Round 1, Cross Validation will identify the part 1 data as a
Test Data Set, and parts 2 – 4 as Training Data Set. In the second round, Cross Validation will indicate
the part 2 data as Test Data Set, and parts 1 and 3 – 4 as Training Data Set. In the third round, Cross
Validation will determine the part 3 as Test Data Set, and Parts 1 - 2 and 4 as Training Data Set. And
finally in the fourth round, Cross Validation will set the part 4 data as Test Data Set, and Part 1 - 3 as
Training Data Set, as seen below.

84

Classification

Figure 2. Cross validation

This processing allows the Classification Model to be taught and tested within the entire dataset. As
a result, even though the data scientists have received the data set with less instances, they can teach the
machine learning very well. To test the Classification Technique accuracy, the data set is divided into k
number of rounds. The mean precision of all rounds is taken to compute in order to obtain a precision
value, the representative of the processing of the data set for each round.

Model Accuracy

Data scientists can measure the efficiency of data classification using the Confusion Matrix which
presents not only the number of the correct classification, but also the number of misclassifications as
seen in the following example (Görtler et al., 2021).

85
 
Classification

Table 1. Confusion matrix

Predicted Value
Yes No
Yes 40 10
Actual Value
No 30 20

According to the table, if the results of the performance, evaluated with the Confusion Matrix, are
obtained as seen in the table, it means that there are 40 “Yes” of correct classification, “True Positive”,
and there are 10 of the misclassifications, “False Negative”. This means there are 50 “Yes” in the actual
data, but the model can correctly recognize only 40. On the other hand, there are 30 “No” of the correct
classification, which is then called “False Positive”, and there are 20 of data misclassification, which
is called “True Negative”. It means that in the actual data, there are 50 “No”, but the model is able to
correctly recognize only 20.

Precision and Recall

To assess the data accuracy, precision can also be used to show the accuracy of classification for data
such as Soaps or Shampoos. Recall is to present the accuracy of classification according to the actual
data, such as Soaps and Shampoos as seen in the following example (Khan, & Ali Rana, 2019).

Table 2. Precision and recall

Predicted Value
Recall
Soap Shampoos
Soap 40 10 80%
Actual Value
Shampoos 30 20 40%
Precision 57% 66%

From the table, data scientists can determine the accuracy of the data classification with 2 measure-
ments. First is Precision. It shows the accuracy ratio of each class as Soap has 57% accuracy. It is cal-
culated by taking 40 to divide by 70. The total classification of the Soap and Shampoo data shows the
accuracy of 66%, which is calculated by dividing 20 by 30, which is the proportion of all classifications
of Shampoos. The other measurement is Recall, the real data that is accurately classified. The accuracy
corresponds to the actual data of Soap is 80%, which is calculated by dividing 40 by 50. And the accuracy
of the actual Shampoo data is 40 percent, which is calculated by 20 divided by 50.

86

Classification

OVERFITTING AND UNDERFITTING

In processing with data mining techniques, data scientists can encounter problems with the outcomes
in two ways:

1. Overfitting: When the data set used to process and the result share a very close relationship, the
results obtained from forecasting or classification are very accurate (Junaidi et al., 2021). It was,
however, found that the results of forecasting or classification show very low accuracy when testing
the model.
2. Underfitting: When the results from teaching machines and the model test have very low accuracy,
the obtained results, when tested, shows low accuracy (Zhang, Zhang, & Jiang, 2019).

To figure out these two issues, the scientists need to use the Cross Validation technique so that the
model can learn the entire dataset. And the performance of data forecasting or classifying can be tested
until the problems regarding both overfitting and underfitting can be eliminated. Then, the scientists can
apply that processing to the test which has higher accuracy.

INCREASING ACCURACY OF MODEL DATA CLASSIFICATIONS

For Data Pre-Processing, data scientists will do Data Exploration to explore the outlier data and ma-
nipulate them by considering cutting them along with the Domain Expert’s advice (Song et al,. 2021).
Missing Value data are explored in order to fill them with other values. If there is a small number of
missing values, some missing values are considered being eliminated. Or if there is a large number of
missing values, it should be considering not using the dataset. Attributes that are not useful for data
analysis, such as ID card numbers are considered too so that the dataset has only the attributes that can
be used to analyze the data accurately. Interchangeable attributes such as gender titles, Mr. or Miss, and
male and female are also considered. These two types of data categories can be used interchangeably in
order to create high-precision modeling and reduce too many computing resources. Data scientists can
consider choosing either of them to analyze. Generalization is the arrangement of data into a suitable
format such as grouping revenue into continuous data, ranging 15,000 – 20,000 baht, so that the data
are suitable for the classification techniques.

K-NEAREST NEIGHBOR (K-NN)

K-Nearest Neighbor (k-NN) is a data mining technique in terms of Classification with the characteristics
of Supervised-Learning, which is used to divide the data into 2 parts; to teach the machine learning and
to test the data classification (Rahman et al., 2021) The k-NN also aims to classify the data. For example,
data scientists have data to teach machines to learn about each type of film, such as drama or comedy.
The scientists, then, bring the test data into the model to allow k-NN to determine what form the film
matches. The process is illustrated as follows (Nuttavut Thongjor, 2017).

87
 
Classification

Table 3. Data used for teaching machine learning before processing with k-NN

Movies Crying Scenes Comedy Scenes Types

Ant’s Tear 10 2 Drama
The Mourning Land 8 3 Drama
The Slavery 9 5 Drama
Happy Lettuce 2 15 Comedy
Cat Adventure 3 13 Comedy
Golden Age Amour 9 4 Drama

On the tables, data scientists can input data into models to teach machine learning about movie genres
in order to classify data in the film “Golden Age Amour” by calculating the distance of the film and
other movies. Then, the k-NN technique will classify “Golden Age Amour” in the same category as the
movies with similar distances. It can be calculated according to the following formula

Distance = (Crying Scenes − the DataClassification  rying Scenes ) +(Comedy Scene − the DataClassification Scenes )
2 2
 requiredC  requiredComedy


Data scientists are able to calculate the distances from “Golden Age Amour” to the other films and
shows in the following results:

Table 4. the distances of “Golden Age Amour”

Movies Crying Scenes Comedy Scenes Distance Types

Ant’s Tear 10 2 2.23 Drama
The Mourning Land 8 3 1.41 Drama
The Slavery 9 5 1 Drama
Happy Lettuce 2 15 13.03 Comedy
Cat Adventure 3 13 10.81 Comedy
Golden Age Amour 9 4 ?

From the table, when k equals 3, it means there are 3 films with similar distances to “Golden Age
Amour”, which are Ant’s Tear, The Mourning Land and The Slavery. Golden Age Amour, when processed
with k-NN technique, is therefore classified as a drama. The scientists can use the k-NN technique to
classify data by calculating the nearest distance in order to classify the data from data similarity. The
number of k can determine the amount of data used to consider the similarity.

88

Classification

DECISION TREE

Decision Tree is used for data classification (Puspitasari et al., 2021) The results come in the form of a
tree diagram. Data scientists have to divide the data into two parts, the data for teaching machine learn-
ing and the data for testing. The tree is formed from the results of each test attribute that data scientists
use as the condition for data analysis. The tree diagram shows the results by branching in relation to the
results in the root analysis, where the leaf node of each branch is the classes or the results. To use the
results obtained from Decision Tree analysis, data scientists can take into account the decision-making
conditions.

Figure 3. Data classification with decision tree

89
 
Classification

In the diagram, data scientists create a Decision Tree from the attributes consisting of Refund which
means the tax refund, Marital Status which is the tax payer marital status, and Taxable Income which is
the amount of income that has reached tax threshold. Cheat means to enter false data in order to receive
a tax refund. Each attribute contains the following data.

Table 5. Examples of data used in determining the classification

Refund Marital Status Taxable Income Cheat

No Married 499$ ?

From the sample data, data scientists need to know whether the instance is from someone who falsifies
the data in order to obtain a tax refund. It can be determined from the Decision Tree, that the person who
has requested a tax refund is not eligible for a tax refund if the person is married and has taxable income
less than $500. The instances can tell whether someone falsifies the data in order to obtain a tax refund.

Creating a Decision Tree

Creating a decision tree begins when data scientists consult a Knowledge Expert on which attributes
should be used in the analysis. The complexity of the decision tree should be taken into account. There-
fore, the scientists should not choose a large number of attributes, which would result in the complexity
of data classification. In addition, data scientists can also select attributes by using Information Gain,
which can be calculated in 2 ways: 1. Reduction of Impurity Measure is the value that each class that
can divide into classes. The better the class can be divided, the better the Impurity Measure value will
be lower. 2.Entropy is data uncertainty. The better the ability to divide a class, the lower the Entropy.
To select attributes to create a decision tree, it can split the decision path in order that data scientists
can consider the classification of data in some ways. This book will calculate Information Gain from
Entropy as shown in the example below.
There are totally 14 instances, with the label being voluntary resignation divided into 5 times of no
resignation, and 9 times of resignation. Data scientists can calculate Entropy by the following formula.

Entropy ([9,5]) = - (5/14) log2 (5/14) – (9/14) log2 (9/14)

= 0.94

From the calculations, data Scientists have obtained the Entropy of 0.94, which is very high. It means
that the Information Gain is at the level that should be improved for more capabilities in data classifica-
tion. Therefore, data scientists can calculate the Entropy of each Attribute to compare data classification
performance along with labels, starting from an attribute named high, moderate and low Work Experi-
ence. The data in each category is related to the voluntary resignation information as shown in Table 7.
According to data relationship, scientists must calculate the Entropy for each type of data as follows.

Entropy (High Work Experience, 2 resign, 3 no resign) = – (2/5) log2(2/5) - (3/5) log2(3/5)
= 0.97

90

Classification

Table 6. Examples of information gain data sets

Work Experience Education Family Status Marital Status Voluntary Resignation

High Ph. D. High Single No resign
High Ph. D. High Married No resign
Moderate Ph. D. Moderate Single Resign
Low Master High Single Resign
Low Bachelor High Single Resign
Low Bachelor Moderate Married No resign
Moderate Bachelor Moderate Married Resign
High Master High Single No resign
High Bachelor Moderate Single Resign
Low Master Moderate Single Resign
High Master Moderate Married Resign
Moderate Master High Married Resign
Moderate Ph. D. Moderate Single Resign
High Master High Married No resign

Table 7. Relationship of work experience data with voluntary resignation data

High Work
Resign Resign No Resign No Resign No Resign
Experience
Moderate Work
Resign Resign Resign Resign -
Experience
Low Work
Resign Resign Resign No Resign No Resign
Experience

In addition, when calculated with the same method for data of Moderate Experience and Low Expe-
rience, the entropy values are obtained as 0 and 0.97, respectively. The entropy of the three data types
can be calculated for the average for the labeled data as follows:

Average Entropy = (5/14)0.97 + (4/14)0 + (5/14)0.97

= 0.693

Data scientists obtain the Entropy value of the label classification from an attribute named Experience
at 0.693. The scientists can do this calculation on the rest of the attributes, and then the entropy of each
attribute is subtracted from the total entropy of the data set, in order to obtain Information Gain. For
example, if the total Entropy of this data set is 0.94 – 0.693, which is the Entropy of the Work experience
attribute, the Information Gain value is 0.247. Therefore, when calculating Information Gain together
with other Attributes that have been computed for Entropy, such datasets have Information Gain values​​
for each Attribute as presented in Table 8.

91
 
Classification

Table 8. Information gain values from the sample dataset

Work Experience 0.247

Education 0.029
Family Status 0.152
Marital Status 0.048

Based on the Information Gain value, data scientists can select the Work Experience Attribute, which
has the greatest Information Gain value as the first node, to be able to classify the data effectively in
either direction.

Pruning the Tree

In cases data scientists use a dataset consisting of a large number of attributes, which may or may not be
related. This results in the creation of a large decision tree which contains a large number of Branches
and Leaf Nodes (Karthik & Krishnan, 2021). Moreover, when the data are tested on the model, it has
low accuracy that causes overfitting. Therefore, it is necessary for the scientists to prune the tree in
order to create the decision tree from the attributes useful for decision-making and small enough to use
as a decision-making criterion. Data scientists can determine the highest level of Information Gain that
they can tolerate, so that the level of leaf node within the Decision Tree has appropriate number with-
out the complexity. In other cases, the scientists can determine the minimum number of Branches and
Leaf Nodes by defining values in both the Information Gain form and the number of Branches and Leaf
Nodes. The scientists can define parameters within RapidMiner software, which will be later presented
in the next section.

Strengths and Weaknesses of Decision Tree

Decision Tree is a data mining technique used for data classification. It is easy to understand and suit-
able for analyzing small or Low Dimensionality to enable easy-to-classify decision trees. Data scientists
can select only the attributes that affect the consideration with Information Gain analysis or consulting
with Domain Expert to obtain reasonable classification conditions. However, if the scientists choose a
large number of attributes to create a decision tree, it will result in a large number of leaf nodes and a
complexity in classifying the data. Also, if the inappropriate attributes are selected, it will result in a
classification error.

NAÏVE BAYES

Naïve Bayes is a data mining technique based on Bayesian Classifier, using probability to classify data
within the entire dataset (Rabiul Alam et al., 2021) For example, if we have seen a patient with symptoms
similar to COVID-19 infection, we can use probabilities to calculate the percentage of the possibility
of COVID-19 infection. The main principle of Naïve Bayes is data analysis based on the principle of
probability. Data scientists can classify the data from existing data sets by importing the data sets into the

92

Classification

Table 9. The instances the scientists need to classify with naïve Bayes

Work Experience Education Family Status Marital Status Voluntary Resignation

High Master High Married ?

Naïve Bayes model to teach the machine learning. Then, the model is tested and the data are categorized.
This book uses an example to analyze whether the data sets, including work experience, education, fam-
ily base, status and voluntary resignation, can tell the employees’ voluntariness to resign. The formula
based on Bayes Theorem is presented as follows:

P refers to Possibility
C refers to Data Class
A refers to Attribute

P (A|C ) P (C )
P (C|A) =
P (A)

From the formula, data scientists can classify the data according to the instances as seen in the fol-
lowing example.
From Table 10, the sample datasets can be taken from Table 7 to analyze the probability whether
employees with such data will decide to resign or not.

P (resign| E) = P (work experience= high | resign) x

P (education = master | resign) x
P (family status = high | resign) x
P (marital status = married | resign) x

2 3 3 3 9
* * * *
P (resign) /P (E) = 9 9 9 9 14
P (E )

= 0.0053/P(E)

P (no resign| E) = P (work experience = high | no resign) x

P (education = master | no resign) x
P (family status = high | no resign) x
P (marital status = married | no resign) x

93
 
Classification

3 1 4 3 5
* * * *
P (no resign) /P (E) = 5 5 5 5 14
P (E )

= 0.0206/P (E)

As in the example, data scientists use the datasets from Table 7 to classify the data, divided into 9
employees who voluntarily resigned and 5 employees who voluntarily did not resign out of the total of
14. Therefore, the scientists can use Bayes Theorem to analyze the probability according to Table 27
to classify whether such employees will resign or not. The probability for resignation is analyzed as
0.0053, and the probability of not leaving is 0.0206. Therefore, data scientist could choose the highest
probability level, which is no resignation. The scientists can apply Naïve Bayes to classify data by rely-
ing on the data set imported into the model. The results obtained from the Naïve Bayes model is in the
form of probabilities, which can be used to classify the data.

OTHER METHODS FOR CLASSIFICATION

In addition to the classification techniques discussed in Sections 1–4, data scientists can use other models
to perform classification.
Support Vector Machine (SVM): It is a coefficient calculation used to classify data. The coeffi-
cients are created as a line in the data classification (Hyperplane). Each calculation can create multiple
lines to classify data. But a good line to be used to classify the data must have the nearest distance, from
the line to the data. So, they can split the data appropriately. The classified data are then called Support
Vector (Jamalludin et al, 2021).
Artificial Neuron Network: ANN is one of the Classification Techniques by which data scientists
can divide the data sets into 2; one is for teaching the machines to learn about the data types and datasets,
and the other one is used to test whether the ANN model is able to classify the data precisely. The ANN
technique also optimizes the data to be factual by adjusting the weight of each node within the model.
The accuracy of ANN classification is high (Fadlil, Umar, & Gustina, 2019).
Logistic Regression: According to the concept of Linear Regression, if data scientists need to
forecast data with results that are split into two values, or other results that are Category Data, the data
scientists need to use Logistic Regression to predict it. Therefore, Logistic Regression is used to clas-
sify data by forecasting the likelihood of an outcome as classified (Thangakumar & Kommina, 2020).
For example, the prognosis of a patient’s chances of COVID-19 infection or not is similar to the other
classification techniques. Data scientists need to consult a Domain Expert to select attributes that can
be used to classify data effectively.

DATA CLASSIFICATION WITH RAPIDMINER SOFTWARE

The data classification experiment using RapidMiner software covers three methods: k-NN, Decision
Tree and Naïve Bayes:

94

Classification

k-NN Data Mining Technique

Data scientists can perform k-NN data mining classification experiments using Golf datasets within
RapidMiner.

Figure 4. Sample dataset: Golf

By bringing a sample dataset named Golf into that model, within the datasets are the attributes “Play”
which stands for playing golf, “Outlook” for weather, “Temperature”, “Humanity”, and “Wind”. These
attributes can be used to classify the data and tell whether to play golf or not to play golf. Within the
data set, there are instances as follows:

Figure 5. Details of golf dataset

95
 
Classification

Then the data scientist can right click, select Insert Building Block and select Nominal Cross Valida-
tion to divide the data into Train Set and Test Set.

Figure 6. Using nominal cross validation

Figure 7. Bringing cross-validation into the model

96

Classification

Then, data scientists make a connection between the Golf dataset and Validation, and connect the
results of machine learning to the results of model performance testing. Within Validation, the Decision
Tree data mining technique is set as default.

Figure 8. Default validation

Meanwhile, the data scientists replace the Decision Tree model with the k-NN model as follows.

Figure 9. Using the k-NN technique for data classification

To use the k-NN model, data scientists are required to determine the k number, or the number of
distance analysis. From the example, k is set as 3.

97
 
Classification

Figure 10. Setting k number

Afterwards, data scientists are able to perform data processing experiments with the k-NN data min-
ing technique and then obtain the following results.

Figure 11. Results of data processing with k-NN data mining technique

In the figure, it shows RapidMiner software renders data using precision and recall techniques. The
scientists are able to determine the accuracy of each class. The data on Not Playing Golf shows 20 percent
accuracy, and Playing Golf is 77.8% accurate. Accuracy of classified data of Not Playing Golf is 33.33
percent, and 63.64 percent of Playing Golf. Overall, the accuracy is 55 percent due to the relatively small
amount of data within the dataset.

Data Classification Using Decision Tree

Based on the previous datasets which are Golf and Cross Validation, when data scientists need to analyze
data using Decision Tree, they can use Nominal Cross Validation, which default is set to the Decision
Tree as seen below.

98

Classification

Figure 12. Using the decision tree defaulted to nominal cross validation

Figure 13. Adjusting parameters to create decision tree model

Data scientists can adjust the Criterion’s parameter to Information_gain in order to select the most
suitable attributes to create a Decision Tree. The result of processing is as follows.

99
 
Classification

Figure 14. Results from decision tree processing

Data scientists can use the results of Decision Tree processing to determine the data, by deciding
whether to play golf or not.

Data Classification Using Naïve Bayes

After data scientists have conducted data classification experiments with k-NN and Decision Tree,
another technique that can be used to classify data with RapidMiner software is Naïve Bayes, by using
the Golf and Nominal Cross Validation datasets as seen below.

Figure 15. Data classification using naïve Bayes

100

Classification

After processing the data with the Naïve Bayes technique, the results of the classification of the data
are obtained as shown in Figure 16.

Figure 16. Results from naïve Bayes processing

Data scientists can classify the data by the probabilities of each class. In this experiment, the data
analysis accuracy is 50.0.

CASE STUDY

Comparison of Data Classification Efficiency to Analyze Risk

Factors That Affect the Occurrence of Hyperthyroid Using Data
Mining Techniques (Hongboonmee & Trepanichkul, 2019)

Classification Technique is a Data Mining Technique widely used in many industries. There have been
many theories developed to support the principles of data analysis with such techniques such as ANN or
Decision Tree. In the medical industry, there are many transactions involving patients. Moreover, healthcare
professionals are also keen on using data mining techniques for diagnosis and treatment as well as for
research purposes. For example, data scientists, who may be hospital personnels or doctors themselves,
can collect data on people with diabetes, then use Classification Technique, the Supervised-Learning,
to train machines to learn and be able to identify whether patients are likely to develop diabetes or not.
In 2019, there was a study on a trial comparing the effectiveness of data analysis to classify hyper-
thyroid disease (Hongboonmee & Trepanichkul, 2019). Data were collected from 323 patients from a
hospital in Thailand. The experiment was conducted with tClassification techniques, consisting of artificial
neural networks, naive bayes and decision tree. The purpose of this research is to create native apps to
be installed on smart phones in which users can fill data. It can then analyze their risk of Hyperthyroid.
The data set used to teach machine learning consists of features such as gender, age, fatigue, building
palpitations, mood swings, normal weight gain or loss, sweating, physical weakness, trembling hands,
insomnia, bulging eyes, and the neck size. The Data Object in each feature has the characteristics of
Boolean, Yes and No, and the result of the Hyperthyroid Disease Risk Analysis is a Label, which contains
the Data Object on Risk Level in Numerical Data.
Such an experiment Datasets were taken into Train model, which consists of artificial neural networks,
naive bayes and decision trees, and then it was found that the most effective technique was artificial neural
networks. The research also tested on what features influence on the risk of having Hyperthyroid. If the

101
 
Classification

model’s accuracy is reduced, then that feature does not affect the classification of risks. It also found that
the features influencing the risk of Hyperthyroid are gender, mood swings, and physical exhaustion. The
feature age shows the least influence on model performance, which means that Hyperthyroid disease can
occur in people of any age. Then, the researchers brought the Classification Model to develop Native
Apps, installed on the smart phone, and then the performance of the system is tested among the users
in order to reflect the classification performance of the installed native apps.
From such research, it is obvious that The Classification Technique can be used for classifying data
and responding to questions. It is also imperative that data scientists test a variety of datasets that have
been compiled with various Data Mining Technique in order to test the analytical efficiency before
analyzing the data and interpreting the results. The research also shows that data scientists can perform
individual data analyzes, data analysis using Python, R or even RapidMiner, or deploying the model
from the experiment to allow data users to access data analysis through various applications.

Breast Cancer Prediction and Detection Using Data Mining

Classification Algorithms: A Comparative Study (Keles, 2019)

Breast cancer has long been a deadly disease. The medical industry is trying to spread information
about breast cancer self-diagnosis so that patients can be quickly identified. Early illness of cancers, if
discovered early, it can be cured with a success rate of 90–95%.
In 2019, there was a Breast Cancer Prediction and Detection Using Data Mining Classification
Algorithms: A Comparative Study (Keles, 2019) using a variety of data mining techniques to test the
effectiveness of breast cancer identification. This research reviewed the literature and found that many
other studies have attempted to create data analysis models to identify cancer, tumor and normal tissues
using the Decision Tree, the Neural Network and the Decision Tree and k-NN. The research used a total
of 6,066 instances of datasets related to past breast cancer diagnoses. 5,405 instances were used to train
the model to learn, and 601 instances were used for testing the performance, as well as to optimize the
model training. The study also used the Cross Validation Technique, which divides the dataset into 10
parts each of which is applied to train and test models. This allows the model to be trained from 10
datasets and tested by 10 datasets to bypass the constraints of that dataset, which has too few instances
of data in train and test models.
From the experiment, the top five data mining techniques that perform best in diagnosing cancer
biopsies are Random Forest, Random Committee, Bagging, SimpleCART, and IBK.
In Case Study 1 and 2, the readers will find that even after the scientists have been asked for data
analysis and can analyze the suitable data mining techniques, the Classification Technique, the scientists
have a large number of algorithms within the Data Mining Technique that can be used in response to
data mining, classification or even other types of data mining techniques. Therefore, a method that data
scientists can use is to bring up those techniques and experiments to find the most effective techniques.
However, the fact that such techniques are highly effective at analyzing data at a time does not mean
they are always the most efficient. They vary depending on the datasets that the data scientists teach
the machines to perform. The greater the amount of data a machine learns, the more likely it is to learn
and this will result in a more efficient machine’s ability to classify data, resulting from more learning
in the same way.

102

Classification

Analysis of Work Suitable for Personality With

Classification Technique (Sharma et al., 2021)

Classification Technique can be used in disease diagnosis and applied to analyze data to answer other
questions. In 2021, there is a study called Data Mining Classification Techniques to Assign Individual
Personality Type and Predict. Job Profiles (Sharma et al., 2021) which conducted an experiment on analyz-
ing jobs that match each human personality type using Classification techniques such as Random Forest.
In many situations, people often cannot find work needs that are appropriate for their personality.
The research reviewed the literature on Classification techniques that may be used to classify jobs that
fit the personality of the target audience. The techniques were Logistic Regression, KNN, Support
Vector Machine, Naïve Bayes Classifier, Decision Tree Classifier and Random Forest Classifier. Then,
the most suitable algorithms for processing were compared and a random forest was selected. This re-
search then defines the features that can be used for personality analysis, consisting of living life with
internal data objects which are extraverted and introverted; decision making with data objects which are
planning or using traits; intuition Actions with data Objects which are judgement and empathy. When
analyzing personality, data scientists can analyze all the features into 16 personality types to match the
job characteristics suitable for each personality type. For example, a planner personality is suitable for
a management position.
To create a processing system that can match the jobs with the personality of each user, this research
uses the Django Framework to develop a system in the form of a web application using Python to write
a set of instructions for processing data using the Random Forest technique.
From the case study, the reader can see that besides using Classification Technique to process data-
sets to answer case-by-case questions, data-processing instructions can be embedded into applications
in the same manners as the case studies have used the Django Framework to develop Web Applications,
capable of processing with data mining techniques, Classification. Therefore, for young data scientists
who are studying software used for processing Big Data, data scientists should learn more about software
that help develop applications.

Using Classification Technique Together With Intrusion

Detection System Technology to Classify Threats

Classification itself is also applied in other applications such as network security, especially in networks
full of massive amounts of data traversing public infrastructure such as streaming media. In such cases,
the Intrusion Detection System: IDS is used to detect potential threats. The network system can detect
information or behaviors that are abnormal and can then alert Network Technicians of such threats. This
is different from other security solutions such as firewalls that detect threats in a statistic manner. IDS
can detect threats entering the network in real time, consistent with streaming media data.
Application of Classification Technique, combined with real-time detection of incoming threats to
the network with IDS, can enhance the detection of potential threats. Machine learning can learn the
threats entering the network and then detect all information in real time so that IDS can alert Network
Technicians to deal with that threat in a timely manner. By 2020, there was a research called Data mining
Classification Techniques for Intrusion Detection System (Sharma et al., 2020) that studied the conduc-
tivity of Classification Technique applied to increase the efficiency of work for IDS. The researchers
used Principle Component Analysis: PCA technique to arrange the composition of the feature used for

103
 
Classification

the experiment. Data classification was performed using Random Forest, J48 Decision tree, Random
tree and Naïve bayes. The research used a dataset called NSL-KDD, a dataset on network attacks, for
experimentation. The results showed that Random Forest showed the highest accuracy for classifying
attack data.
Back in 2015, there was another study, Efficient Intrusion Detection System Using Stream Data Min-
ing Classification Technique (Desale, Kumathekar & Chavan, 2015). The same dataset, NSL-KDD, was
used in the experiment, using the Hoeffding algorithm, which features a decision tree learning algorithm,
and the Naïve bayes technique for data classification experiments to encourage threat detection of IDS
more efficiently. From the experiment, it was found that the Naïve bayes technique had the highest data
classification accuracy.
Therefore, data scientists can apply Classification techniques not only in the medical or financial
industry, but also in other applications, such as network security, the process of applying different data
science applications. All of the Dataset acquisitions are on the principles, collection and extraction of
datasets, preparing a dataset, selecting a data mining technique to answer the set questions by perform-
ing a variety of experiments and techniques with the highest efficiency and accuracy in data analysis
and interpretation of results.

CONCLUSION

Data mining techniques is Unsupervised-Learning. The scientists need to divide the dataset into two
parts: one for teaching machine learning and the other one for testing the performance of the model. This
book has given examples of 3 data mining techniques, namely k-NN, Decision Tree and Naïve Bayes.
For k-NN, data are classified by calculating the distances between the required data and the data used
in the analysis. It is to classify the type of data to be analyzed with the data near it. The Decision Tree
technique is a diagram based on Information Gain or Attribute Analysis, which is suitable for data clas-
sification that allows data scientists to take the data to classify and consider together with the Decision
Tree. So that they will obtain the appropriate data classification results. The last section the book brings
up is Naïve Bayes. With this technique, the scientists can take the required data to analyze and process
together with the dataset. This is to perform a probabilistic analysis of data classification. Data scientists
can use RapidMiner software, such as k-NN, Decision Tree or Naïve Bayes, for a quick classification by
bringing the desired data set into the model. Then the Cross Validation model is used to divide the data
into 2; for teaching machine learning and for testing the efficiency of data classification.

EXERCISE QUESTIONS

1. Explain the difference between Clustering and Classification.

2. Describe Cross Validation and its application in RapidMiner software.
3. Explain the concept of k-NN.
4. Describe the result consideration from the Decision Tree obtained from the data analysis.
5. Search for the researches using Classification in data analysis, and then make a brief summary.

104

Classification

REFERENCES

Desale, K. S., Kumathekar, C. N., & Chavan, A. P. (2015) Efficient Intrusion Detection System Using
Stream Data Mining Classification Technique. 2015 International Conference on Computing Commu-
nication Control and Automation, 469-473. 10.1109/ICCUBEA.2015.98
Fadlil, A., Umar, R., & Gustina, S. (2019). Mushroom Images Identification Using Order Statistics Fea-
ture Extraction with Artificial Neural Network Classification Technique. Journal of Physics: Conference
Series, 1373(1), 012037. doi:10.1088/1742-6596/1373/1/012037
Görtler, J., Hohman, F., Moritz, D., Wongsuphasawat, K., Ren, D., Nair, R., Kirchner, M., & Patel, K.
(2021). Neo: Generalizing Confusion Matrix Visualization to Hierarchical and Multi-Output Labels.
Academic Press.
Hongboonmee, N., & Trepanichkul, P. (2019). Comparison of Data Classification Efficiency to Analyze
Risk Factors that Affect the Occurrence of Hyperthyroidusing Data Mining Techniques. Mahanakorn
University of Technology (MUT).
Jamalludin, M. D. M,, Fajar Shidik, G., Zainul Fanani, A., Purwanto, & Al Zami, F. (2021). Implemen-
tation of Feature Selection Using Gain Ratio Towards Improved Accuracy of Support Vector Machine
(SVM) on Youtube Comment Classification. 2021 International Seminar on Application for Technology
of Information and Communication (ISemantic), Application for Technology of Information and Com-
munication (ISemantic), 2021 International Seminar On, 28–31.
Junaidi, A., Ferani Tanjung, N. A., Wijayanto, S., Lasama, J., & Iskandar, A. R. (2021). Overfitting
Problem in Images Classification for Egg Incubator Using Convolutional Neural Network. 2021 9th
International Conference on Cyber and IT Service Management (CITSM), Cyber and IT Service Man-
agement (CITSM), 2021 9th International Conference On, 1–7.
Karthik, M. G., & Krishnan, M. B. M. (2021). Detecting Internet of Things Attacks Using Post Prun-
ing Decision Tree-Synthetic Minority Over Sampling Technique. International Journal of Intelligent
Engineering & Systems, 14(4), 105–114. doi:10.22266/ijies2021.0831.10
Keles, M. K. (2019). Breast Cancer Prediction and Detection Using Data Mining Classification Algo-
rithms: A Comparative Study. Tehnicki Vjesnik - Technical Gazette, 26(1), 149.
Khan, S. A., & Ali Rana, Z. (2019). Evaluating Performance of Software Defect Prediction Models
Using Area Under Precision-Recall Curve (AUC-PR). 2019 2nd International Conference on Advance-
ments in Computational Sciences (ICACS), Advancements in Computational Sciences (ICACS), 2019
2nd International Conference On, 1–6.
Liu, L. (2021). Design of Human Resource Management Information System Based on Decision Tree
Algorithm. 2021 Global Reliability and Prognostics and Health Management (PHM-Nanjing), Reliability
and Prognostics and Health Management (PHM-Nanjing), 2021 Global, 1–6.
Mishra, A., & Vats, A. (2021). Supervised machine learning classification algorithms for detection of
fracture location in dissimilar friction stir welded joints. Frattura e Integrita Strutturale, 15(58), 242–253.
doi:10.3221/IGF-ESIS.58.18

105
 
Classification

Mladenova, T. (2021). A Feature-Weighted Rule for the K-Nearest Neighbor. 2021 5th International
Symposium on Multidisciplinary Studies and Innovative Technologies (ISMSIT), Multidisciplinary Stud-
ies and Innovative Technologies (ISMSIT), 2021 5th International Symposium On, 493–497.
Mnich, K., Polewko-Klim, A., Kitlas Golinska, A., Lesinski, W., & Rudnicki, W. R. (2020). Super
Learning with Repeated Cross Validation. 2020 International Conference on Data Mining Workshops
(ICDMW), Data Mining Workshops (ICDMW), 2020 International Conference on, ICDMW, 629–635.
Mohan, L., Jain, S., Suyal, P., & Kumar, A. (2020). Data mining Classification Techniques for Intrusion
Detection System. 2020 12th International Conference on Computational Intelligence and Communica-
tion Networks (CICN), 351-355.
Nuttavut Thongjor. (2017). Understanding the data classification: k-Nearest Neighbours. Retrieved 9
August, 2021, from https://www.babelcoder.com/blog/articles/k-nearest-neighbors
Puspitasari, D. I., Riza Kholdani, A. F., Dharmawati, A., Rosadi, M. E., & Mega Pradnya Dhuhita, W.
(2021). Stroke Disease Analysis and Classification Using Decision Tree and Random Forest Methods.
2021 Sixth International Conference on Informatics and Computing (ICIC), Informatics and Computing
(ICIC), 2021 Sixth International Conference On, 1–4.
Rabiul Alam, M. G., Hussain, S., Islam Mim, M. M., & Islam, M. T. (2021). Telecom Customer Behavior
Analysis Using Naïve Bayes Classifier. 2021 IEEE 4th International Conference on Computer and Com-
munication Engineering Technology (CCET), Computer and Communication Engineering Technology
(CCET), 2021 IEEE 4th International Conference On, 308–312.
Rahman, B., Hendric Spits Warnars, H. L., Subirosa Sabarguna, B., & Budiharto, W. (2021). Heart Dis-
ease Classification Model Using K-Nearest Neighbor Algorithm. 2021 Sixth International Conference
on Informatics and Computing (ICIC), Informatics and Computing (ICIC), 2021 Sixth International
Conference On, 1–4.
Sharma, M., Joshi, S., Sharma, S., Singh, A., & Gupta, R. (2021). Data Mining Classification Tech-
niques to Assign Individual Personality Type and Predict Job Profile. 2021 9th International Conference
on Reliability, Infocom Technologies and Optimization (Trends and Future Directions) (ICRITO), 1-5.
Song, H., Fu, Y., Saket, B., & Stasko, J. (2021). Understanding the Effects of Visualizing Missing Val-
ues on Visual Data Exploration. 2021 IEEE Visualization Conference (VIS), Visualization Conference
(VIS), 2021 IEEE, VIS, 161–165.
Thangakumar, J., & Kommina, S. B. (2020). Ant colony optimization based feature subset selection with
logistic regression classification model for education data mining. International Journal of Advanced
Science and Technology, 29(3), 5821–5834.
Vichi, M., Ritter, G., & Giusti, A. (2013). Classification and data mining. Springer Verlag.
Zhang, H., Zhang, L., & Jiang, Y. (2019). Overfitting and Underfitting Analysis for Deep Learning Based
End-to-end Communication Systems. 2019 11th International Conference on Wireless Communications
and Signal Processing (WCSP), Wireless Communications and Signal Processing (WCSP), 2019 11th
International Conference On, 1–6.

106
 107

Chapter 6
Deep Learning

ABSTRACT
Neural network and deep learning techniques are essential tools for data scientists when analyzing big
data for forecasting and classification. In supervised learning, data sets are divided into training sets
and test sets, and neural network repeatedly adjusts the weight of data to better represent the actual
data. This book offers a practical guide to performing a neural network experiment with RapidMiner,
which readers can follow step-by-step. For big data, especially non-linear data, deep learning can be
employed. This chapter introduces two types of deep learning: convolutional neural networks (CNN)
for picture analysis and recurrent neural networks (RNN) for sequential or time series data. The book
provides a demonstration of both techniques using RapidMiner, making it accessible to readers who
wish to deepen their understanding of these powerful tools.

INTRODUCTION

In the early days when people first mentioned artificial intelligence, there was a lot of motivation to
develop artificial intelligence capable of human-like responses. However, the challenge for the era was
that artificial intelligence computing required high-performance computing technology and large volumes
of experimental data, resulting in the gradual advancement of artificial intelligence. Now in the era of
high-performance computing technologies such as GPUs with a high-speed network (Telikani, Shah-
bahrami, & Gandomi, 2021), data scientists are able to use data mining techniques to operate on high-
performance computing technologies and collect big data for experimentation using techniques that are
suitable for such processing. Deep learning technique is used to analyze unstructured data consisting of
images, sounds, and text (Fernando et al., 2021: Hongyi Zhu, Samtani, Brown, & Hsinchun Chen, 2021).
There are two types of deep learning: Convolutional Neural Networks and Recurrent Neural Networks
(Lakshmi Devi & Samundeeswari V, 2021: Snineh et al., 2021). The deep learning model modulates
the pre-configuration of data for processing in conjunction with neural network techniques. Artificial
Neural Networks (ANNs) is a data mining technique that offers both classification and numerical pre-
dictions, which is considered supervised learning (Thankachan, Prakash & Jothi, 2021). Data scientists
must first teach machines to learn before testing the data. In the case data scientists classify data with

DOI: 10.4018/978-1-6684-4730-7.ch006

Copyright © 2023, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
 
Deep Learning

other data mining techniques, the data are classified by a linear plane. On the contrary, using neural
networks, data scientists can classify data that are closely attached to facts with a Non-Linear Function.
The neural network is, therefore, used in 2 ways: first, Pattern Recognition, such as user face recognition
for identification or authentication for accessing the phone (Ghorpade & Koneru, 2021), and finally
Forecasting such as forecasting the trend of stock prices (Chinnarasri, Nonsawang & Supharatid, 2012).
This book discusses the principles of artificial neural network and deep learning in both CNN and
RNN formats as follows:

ARTIFICIAL NEURAL NETWORK: ANN

Conceptual Work of Artificial Neural Network: ANN

Artificial Neural Networks simulate human brain activities by having each node represent a brain cell,
and connect each node to form a network which will then be used for processing (Chow and Cho, 2007:
Aggarwal, 2018). The neural network therefore consists of 4 parts. 1. Input Neural is used to support
the data which will be analyzed with the model. The number of input nodes depends on the number of
variables or features used to analyze the data. 2. Output Neural is the result obtained from processing.
It can have one or more results depending on data analysis questions. 3. Hidden Layer is each layer of
nodes that enhance the processing efficiency of the neural network, and 4. Links is the connection of
each node. Each Link calculates Weights, which must be passed on to the next node. Therefore, the
neural network optimizes the nodes, so that the scientists can analyze the responses from the dataset
imported into the model. The primary model of the neural network is Perceptron. The model is designed
to analyze data with input neural and output neural without creating hidden layers to predict numbers.
To teach machines learning and to test data with a neural network both focus on adjusting the weights
of each link, known as the Back Propagation Artificial Neural Network (BP-ANN) (Ramirez-Hernandez
et al., 2020: Guo, Zhang and Chen, 2021) with the process as follows.

1. Artificial Neural Network model estimates the analysis results.

2. The results are compared to the actual results that the scientists have taught to the machine learning
in order to analyze Error.
3. The model takes the error to adjust the Weights in each layer in order to achieve a tightness between
the analyzed value and the actual data. This process is repeated until the error values reach the
acceptable level.

Iterating to adjust such Weights is to teach machine learning and increase the accuracy of data analy-
sis. Iterating each time is called Epochs. Data scientists put 1,000 records into a neural network model,
and the number of Epochs is 500, there will be 500,000 times of iteration to teach the machine learning.
Therefore, the greater the number of data and the number of nodes within the model is, the more time
and digital processing the data scientists need. From the working process that requires computation to
analyze all the data, the neural network model is suitable for numerical data. Therefore, if data scientists
need to analyze text data in the data preparation process, it is necessary to convert letters into numbers
before entering the data into the model.

108

Deep Learning

Figure 1. Artificial neural network

DATA ANALYSIS WITHIN ARTIFICIAL NEURAL NETWORK

Calculating the Output of Artificial Neural Network

Within the neural network, two weights are used to adjust the accuracy of the analysis; 1. Weight from
Input and Hidden Layer nodes, and 2. Weight from nodes that act as bias. Within every layer, it consists
of 1 node that is additional bias.
From the figure, the neural network and the output calculation use following formula.

Y =b +W
 1X1 + W2 X 2 ………W
 n X m

where Y refers to Output

b refers to Weight from Bias
Xm refers to the number of Features
Wm refers to Weigh from Feature

According to the formula, the artificial neural network calculates the output values by getting the
sum of all input values multiplied by the weight of each node, and adding them to the bias values for
each layer. The artificial neural network then uses an Aviation function to adjust the Linear data into
Non-Linear data which is between 0 – 1. The data obtained from the computation are closer to the facts
until the computed output is obtained.

109
 
Deep Learning

Figure 2. Components of weight

Data scientists can train computations within the artificial neural network to teach machines to learn,
as in the following example.

Feature A = 0.4 where Weight = 0.3

Feature B = 0.6 where Weight = 0.5
Output = 0.8

From the data calculated with 0.4*0.3+0.6*0.5+0.8 = 1.22, the value is then adjusted to be Non-
Linear by Logistic Activation Function using the formula f(x) =1/(1+e^(-(x))). From the sample data,
it can be calculated as 1(1/+e-1.22) = 0.77.
When comparing the output from the neural network 0.77 with the actual value 0.8, the neural network
re-adjusts the weight to bring the result as close to the facts as possible by performing the calculation.

(k +1) (k )
( )
k
wj = w j +l yi −yi^ x ij

where i is the sequence of the records within the dataset

k is the number of iterations
j is the Feature sequence
(k )
w j is Input of Weight brought into j iteration
yi is the actual Output value of each record

110

Deep Learning

k
yi^ is the predicted value of each record in each iteration cycle
x ij is the Feature initial data in each record
l is the Learning Rate, being between 0 – 1. The closer to 1, the higher learning rate. The closer to
0, the lower the learning rate.

According to the formula, data scientists can adjust the Weight in order to obtain results that are
closer to the actual value as presented in the following example.

Feature A = 0.4 where Weight = 0.3

Feature B = 0.6 where Weight = 0.5
Output = 0.8
Learning Rate = 0.5

From the example, the output is calculated as 0.77 and when compared to the actual value, it is 0.8.
Therefore, the Weight is adjusted by

adjusting Weight for Feature A which is 0.3+0.5*(0.8-0.77) *0.4 = 0.306

adjusting Weight for Feature B which is 0.5+0.5*(0.8-0.77) *0.6 = 0.509
adjusting Weight for Output which is 0.8 + 0.5*(0.8-0.77) = 0.815

The artificial neural network performs calculations to adjust Weight and repeats the number of cycles,
the data scientist has set, until the error is at the required level in order to provide the most accurate data.

Hidden Layer and Stopping Conditions

In case the neural network does not have a hidden layer, the output can be used to adjust the weight of
the input node to make the values obtained from the model as close to the actual value as possible. But
in the case of the artificial neural network has a hidden layer, the value of the output in the next layer
must be adjusted, as shown in the example below.
In the figure, Weight from node D to node F is 0.5, and node F has an error of 1.2. Weight can be
adjusted by 1.2 * 0.5 = 0.6. Weight from node D to node G is 0.2, and node G shows Error value as 0.4.
The Weight can be adjusted by 0.2 * 0.4 = 0.08, so node D can then adjust all Weight by 0.6 + 0.80 =
0.68. After adjusting the Weight through iterations or the error is set in the specified point, Neural net-
work models will then stop computation to teach machines to learn, which is known as Stopping Condi-
tions. Data scientists can perform calculations to obtain the appropriate error level to stop calculations
in weight adjustments. Data scientists can calculate the Total of Sum Square Error by subtracting each
actual output by the value obtained from the forecast. The result is then squared of 2, and the calculated
results for each output are then combined as an error level appropriate for stopping the weight adjustment.

111
 
Deep Learning

Figure 3. Adjusting the hidden layer weight

ANALYZING DATA WITH ARTIFICIAL NEURAL

NETWORK USING RAPIDMINER SOFTWARE

Data scientists can take the desired dataset to analyze both classification and numerical forecasting into
RapidMiner software, and then create the artificial neural network model as seen in the following figure.

Figure 4. Starting neural net

112

Deep Learning

Data scientists can begin using Artificial Neural Network model with the sample process within
RapidMiner software called _NeuralNetwork. The process prepares an entire numerical dataset, and uses
a Neural Net model to extract those data. Data scientists can define parameters as follows:

Figure 5. Parameters of the neural network

From the figure, data scientists can determine the number of hidden layers. The Training Cycle is
then specified to determine the number of times of machine teaching. The learning rate is the machine
learning rate that ranges from 0 to 1, and the Momentum is the indication of how much the Weight cal-
culated in the previous round, which will affect the weight adjustment in the current round. The scientists
will then decide whether to adjust the learning rate over time. If yes, select Decay. Regarding Shuffle,
it is the sequence in which the input data is imported into the new model. About Normalization, it is
to normalize the data between 0 – 1. Finally, Error Epsilon is to determine the model to stop analyzing
when the Error value is less than a predetermined value. After that, the scientists can execute the model.
The results are as follows.

113
 
Deep Learning

Figure 6. Execute results of artificial neural network

In the figure, the input node corresponds to the number of features used in the data analysis. The
sample has 5 features. The neural network generates 5 input nodes, and there are additional bias nodes.
The output node corresponds to the target variable. In the example, there is a feature that represents
1 node target variable. The scientists can adjust and increase the number of hidden layers in order to
increase the analysis complexity for the artificial neural network.

Figure 7. Adding a hidden layer to artificial neural network

114

Deep Learning

Data scientists can press Edit List button to add hidden layers by specifying the number of hidden
nodes. From the figure, the scientists define 2 hidden layers. Layer 1 has 4 hidden nodes, and layer 2 is
determined to have 2 hidden nodes. The result of execution is as follows:

Figure 8. Adding hidden layers

In the figure, when data scientists set the nodes in layer1 as 4. The additional bias nodes are added,
and in layer 2 it is set to have 2 additional bias nodes. When data scientists need to classify the data using
artificial neural networks, they can conduct an experiment using a sample dataset within the RapidMiner
software called Iris, which contains data about the sizes of the Iris flower petals. The experiments can
be carried out to identify the Iris flower species from the data set.

Figure 9. Iris classification experiment

115
 
Deep Learning

In this figure, data scientists take the Iris data into the process, then the Cross Validation model is
used to divide the data into two parts; the data used for teaching machine learning and the data for the Iris
species classification test. The result of these 2 parts, the neural network model and the neural network
model’s classification efficiency, are then connected Within Cross Validation.

Figure 10. Processing results with neural net

In the figure, the Cross Validation model teaches the machine to learn the data about the width and
the length of the Iris petals with the Neural Net model in order to identify Iris flower species, and build
a neural network. Then the Apply Model is used to generate the results of the classification of the Iris
flower based on the width and length of the petals. In the last section, the accuracy of Iris flower species
classification is analyzed.

Figure 11. Neural network for Iris flower species identification

116

Deep Learning

As shown in the figure, the artificial neural network is made up of 4 input features, which are the
width and length of the Iris flower. There is 1 hidden layer consisting of 5 nodes, 3 output nodes (3 Iris
flower lines). The accuracy of data classification is as follows.

Figure 12. Accuracy of data classification with artificial neural network

After analyzing accuracy of Iris data classification, it is found that the artificial neural network has
an accuracy of 97.33%. Data scientists are able to modify the number of hidden layers to increase the
complexity of data processing for the artificial neural network. However, the number of hidden layers
that can be used for high-precision data analysis depends on the nature of the dataset, and the accuracy
does not vary with the increasing number of hidden layers. Therefore, it is necessary that data scientists
prepare the data ready for the analysis, and conduct various data analysis experiments for the most ac-
curate data analysis results.

DEEP LEARNING

Convolutional Neural Networks (CNN)

Convolutional Neural Network (CNN) supports large data analysis, especially image data (Sriwiboon,
2020). Therefore, CNN is used in the work related to the artificial intelligence to analyze image-related
data, such as the route classification of autonomous vehicles through camera images. Knowledge of CNN
starts with the research on cat’s vision, which uses the visual method as parts. These parts are then com-
bined into a whole picture, with the cat’s brain focusing only on each part of the picture. Data scientists
develop data by converting images into numerical format, and then CNN technique is used to classify
images into parts similarly to cat’s brain. If the images have been compiled in the form of numbers and
as features, data scientists can use artificial neural network processing techniques for classification such
as images of cats, dogs or elephants. However, those images need to be modeled. The scientists can use
CNN techniques to prepare those data as in the following process.
Feature Extraction: It is to transform the images into the number format. Data scientists focus on
the features associated with the images; such as the elephant image which has distinctive features such
as ears and ivories. Data scientists use a filter matrix to count whether each part of the image has data
that correspond to the features of interests.
Filter Matrix: It is used to check patterns in each part of the images in order to obtain features. To
check each part of the image, the convolutional operation method is used to obtain the number format
data used to represent each feature. There is a method for calculating according to the example below.

117
 
Deep Learning

In the case that the data scientists receive a 10 x 10 image as input (Original Image), then assigns a
3 x 3 filter matrix size, and set the Filter Matrix to check the format of the image at a time, 1 row and 1
column. Moving the Filter Matrix is called Stride. Sizes of Feature Map can be calculated as = (N - F)
/ Stride + 1.

N = 10 x 10
F=3x3
Stride = 1
Size of Feature Map = (10 - 3) / 1 + 1 = 8

Therefore, after the convolutional operation, the image size will be reduced from 10 x 10 to 8 x 8,
which will store only the necessary parts, such as cat images. The Filter Matrix examines the cat’s eyes
and nose only. After the convolutional operation, it will store only the specific parts of the eyes and
noses, so that it can remember whether the images are cat images.
In this convolution process, the ReLU algorithm is used. When the convolutional operation is nega-
tive, the ReLU algorithm adjusts the value to 0, so that the data is a non-linear function, which is close
to the facts of data classification.
Data scientists use a filter matrix to detect the data patterns, and allow the features to store only the
unique images with Convolutional Operation. The next process is Pooling which is to reduce the size of
the data. Data scientists can use either Max Pooling which is the selection of the largest number in each
image data segment or Average Pooling which is to take all the numbers in each image data segment to
create an average to analyze the data. The example of Max Pooling and Average Pooling is as follows.

Table 1. Sample data set for max pooling and average pooling

4 3 8 7
1 7 8 1
9 6 3 2
3 4 0 4

From the sample dataset, the scientists can do Max Pooling, covering 2 rows and 2 columns, and
select the highest value of Data Objects. The result is as follows.

Table 2. Results from max pooling

7 8
9 4

118

Deep Learning

From the Max Pooling table, the size of the dataset can be reduced by selecting only the largest value
each time. And based on the results of Max Pooling, the data scientists can also choose to use Average
Pooling from the same dataset.

Table 3. Results of average pooling

3.75 6
5.5 2.25

Feature Extraction, therefore, uses the algorithm of Convolutional Operation along with Pooling
to extract image-specific data before transferring it to the artificial neural network for classification as
shown in the figure.

Figure 13. CNN performance

From the picture, the image data is verified and converted into the form that can be used for processing
through the feature extraction process. To do this, v Convolution along with Pooling is used to collect
only the distinctive features of the image. Then flattening is done to adjust the format of the matrix data
into vector. This process is called Fully Connected, which is to put the data parts together in a feature,
ready to use in classification process with the artificial neural network. When required to classify data,
the scientists need to consult with an image related professional to determine which distinctive feature
can be used to classify data from the images. The size of the Filter Matrix, the data format and Stride
frequencies are to be considered in order to design the resolution of data extraction appropriate for the

119
 
Deep Learning

questions. When considering the CNN technique suitability, it is suitable for analyzing unstructured data
such as images or sounds. For example, it is used to classify cancer-causing tumors from X-Ray images.
With the CNN technique that analyzes the data by extracting the data into parts, the result is that each
data part shows no relation. Therefore, the data scientists focus on analyzing the relationship of the data,
CNN techniques may not be suitable for such problems. The scientists need to provide digital resources
with high-performance computing and large datasets to teach machines to learn. This amount of data
affects the efficiency of the model’s classification.

RECURRENT NEURAL NETWORKS (RNN)

In the case that the problems focus on analyzing the relationships of Sequential Data and Time Series
(Sabau et al., 2021) such as Language Translation, Natural Language Processing (NLP) and Image
Captioning in daily life, RNN is used in the Search Engine. When the user search for a word, the search
engine will suggest the texts related to the text. The RNN model has taught the machine to learn from
search queries entered by users around the world until it can analyze the relationship of the data used by
users who search through the search engine.
The main characteristics of RNN include: 1. The ability to support data import with the inequitable
data size into the neural network, 2. The ability to verify the relationship between data objects, and 3.
The ability to arrange the data in sequence such as the arrangement of the sentence structure by RNN
processing. The RNN architecture is as follows.

Figure 14. RNN architecture

In the figure, the RNN processes the data to analyze the relationship between the data objects. When
the data is imported into the network (x), the RNN technique calculates the resulting value (y) by tak-
ing the input from the previous step and coprocessing it. It is calculated within the network as follows:

120

Deep Learning

Yt = Whyht
Ht = f (ht-1, xt)
Ht = tanh (Whhht-1+WxhXt)

The y value, or the result in the current step, is calculated from the Why, which is the Weight of the
node multiplied by the ht value. The ht value is derived from ht-1, which is the value from the previous
step. The xt value is the input at the current stage. The tanh function is then used to convert the value
from – 1 to 1, resulting in a non-linear function. Thus, the RNN algorithm can analyze the relationship
of the data by combining the values from the previous data, and analyzing it until the result has become
continuity of correlated data.
However, when the data set is imported into a model for correlation analysis, if the correlated data
objects have a large storage distance, this results in discrepancies in the analysis of the sequence arrange-
ment. Therefore, data scientists can use the Long Short-Term Memory (LSTM) which has the ability to
memorize values from the previous step, that are very distant from the current node, to compute with
the current node in order to obtain the result from proper sequence arrangement.

USING CNN DEEP LEARNING TECHNIQUES WITH RAPIDMINER SOFTWARE

To use deep learning techniques with RapidMiner, though within the software has prepared the basic
models such as Deep Learning H2O, data scientists are required to install the Deep Learning ND4J
extension in RapidMiner when they want to process CNN and RNN models.

Figure 15. Installing extensions

As shown in the figure, to install extensions, data scientists can select the Extensions menu and then
select Marketplace (Updates and Extensions).
When entering the RapidMiner Marketplace, the scientists then search for ND4J and press Search.
When seeing the extensions related to the machine learning model, then, select all 3 extensions namely
ND4J, Deep Learning and Image Handling by clicking on the extension selecting for Installation and
press Install Packages. After completing the installation process, data scientists need to restart the
RapidMiner software.
To develop models with CNN deep learning techniques, this book has used a sample image data-
set from website https://storage.googleapis.com/tensorflow/tf-keras-datasets/mnist.npz. The dataset is
MNIST – JPG - Training, which contains a total of 10,000 instances (Beohar and Rasool, 2021: LeCun,
Cortes and Burges, 2021) as seen in the example.

121
 
Deep Learning

Figure 16. Searching for extensions and installing it

Figure 17. Sample data within the experimental dataset

From the image data set, data scientists can use RapidMiner software to model CNN machine learn-
ing. The process is as follows.
In the figure, the data scientists have brought the MNIST – JPG – Training dataset into the model
using Read Image Meta Data, and then split the data into two parts with Split Data. The data are divided
into 2 parts; 80% for teaching machines, and 20 percent for testing the accuracy of the model. Then,
the data for the machine learning is imported into the pre-process image to be transformed into Tensor
object in the form of vector, in such a way that Deep Learning (Tensor) model can process. The data
from the Pre-Process Image and from the Deep Learning (Tensor) are then entered into the Group Model
(Generic), which enables the Pre-Process Image to be co-processed with the Deep Learning (Tensor).
The accuracy of the processing is then proven with the Apply Model (Generic). The Append model is
then used to compare the datasets from forecasting before sending them to the Performance in order to
show the accuracy of all data classification processing.

122

Deep Learning

Figure 18. Model of CNN learning techniques

Within Deep Learning (Tensor), data scientists can use models to define CNN-based processing
consisting of Convolutional Operation, Pooling and Fully-Connected as seen in Figure 19.
From the figure, data scientists can define each section of the parameters to determine CNN process-
ing as shown in Figure 20.

Figure 19. Setting convolutional operator model, pooling, and fully-connected

Figure 20. Setting parameters to add convolutional layer

123
 
Deep Learning

Data scientists determine the Size of Feature Map to a frequency of 12, covering the important area
on the image by setting the Number of Activation Maps to 12. The size of the Filter Matrix can be set
as 5 x 5, and the Filter Matrix move is also set in order to detect every 1 row and 1 column by setting
the Stride Size to 1 x 1.
In Figure 21, the scientists can select the method to reduce the data size by setting the Pooling Method
to Max, then scale the Filter Matrix by setting the Kernel Size to 2 x 2, and shift the Filter Matrix for
detection every 1 row and 1 column with Stride Size of 1 x 1.

Figure 21. Setting parameters to add pooling layer

Figure 22. Setting parameters to add fully-connected layer

124

Deep Learning

When classifying the data, the number of Neurons must be equal to the number of data types. In
this case, the image data is 0–9. So, if there are 10 data types in total, the data scientists have to set the
number of Neurons to 10 and the Activation Function into Softmax format by setting Parameter to add
Fully-Connected Layer. It is to adjust the data into feature forms that can be used for processing with
the artificial neural networks.
In each part of the model, data scientists can define parameters so that the model performs the pro-
cessing appropriate for the questions.

Figure 23. Setting parameters to read image meta data

Data scientists can use Read Image Meta Data to bring image data into models by setting the Direc-
tory to point to the data source by selecting the folder button, and accessing the source address of the
dataset. Then the target variables are set by selecting Use Label. The data set used in this experiment has
0-9 Sub Folders, so the target variables for each Sub Folder is then set according to the folder name, 0-9.
Data scientists can divide data into two parts; one for teaching machine learning, and the other for
testing; by setting in the Partitions section, and then pressing the Edits Enumeration button.
Data scientists can determine the proportions of the data by selecting Add Entry, then setting Part
1 to 0.8 and Part 2 to 0.2. After that, the scientists can determine parameters to Pre-Process Images by
setting Path as Path.
Then, data scientists can define a model within Pre-Process Images in order to manipulate with im-
age data as follows:
Meanwhile, a parameter for Deep Learning (Tensor) is determined, which also processes the data
with Convolutional Operation, Pooling and Fully-Connected.
In the figure, data scientists define Epochs as 30, which is the number that the data set teaches the
machine to learn. Then, within Deep Learning (Tensor), the structure of the processing can be defined
as follows.

125
 
Deep Learning

Figure 24. Setting parameters to split data

Figure 25. Partition setting to divide the dataset

After the scientists have defined the parameters to the Add Convolutional Layer, Add Pooling Layer,
and Add Fully-Connected Layer as in the previous step, the data scientist can add the Add Fully-Connected
Layer before the last step. And the number of Neurons is set to 100 to reduce the data from the Add

126

Deep Learning

Pooling Layer before sending it to Add Fully-Connected. The number of Neurons is set as 10 according
to the number of data types which is 0 – 9. Once a deep learning model has been created, and param-
eters have been assigned to every model, data scientists can execute the model. The result is as follows.

Figure 26. Setting parameters to pre-process images

Figure 27. Path setting within pre-process images

Processing results show a comparison between facts and results obtained by classification using a
CNN-type deep learning model. For example, the real data is the number 7, and the result of the clas-
sification of the data is the number 7. However, this experiment has used a data set with only 10,000
instances; 8,000 instances to teach machine learning, and 2,000 instances to test the model. It results in
10.30 percent of teaching machine learning efficiency, and the numbers that have been correctly classi-
fied is only 7 out of 10 categories.

127
 
Deep Learning

Figure 28. Setting parameters to deep learning (tensor)

Figure 29. Determining the number of neural for each add full-connected layer

Figure 30. Processing result

128

Deep Learning

Figure 31. Accuracy of data processing

CNN type deep learning techniques are suitable for analyzing big data for classification and teach-
ing machine learning. Therefore, data scientists need to use GPU Processing to reduce the time spent
on processing. Moreover, processing a large number of instances, and using a large amount of data can
promote more accuracy in a CNN-type deep learning model.

THE USE OF RNN DEEP LEARNING TECHNIQUES

WITH RAPIDMINER SOFTWARE

Once data scientists encounter difficulties analyzing data to predict outcomes from time series data,
they can use RNN. In this paper, the following samples are taken from RapidMiner software for testing.
The sample model is a forecast of oil prices. Within the data set, the features are the sequence of oil
price changes, rate of change, date of change and oil prices. There are 10,000 instances. Daily oil price
and each sequence of the change depend on the price of the previous day.
Based on the datasets, data scientists can use RNN to forecast oil prices from the continuous and
correlated data. The models can be created according to the example.
From the sample model, the Gas Station dataset is imported into the model. Then, the nominal data
are changed to numerical with Nominal to Numerical, so that they can be processed by artificial neural
network techniques. The data are then divided into two; one for teaching machine learning, and the other
for testing the performance of the model with the Filter Example Range. Then the Normalize model is
used to adjust the numbers in the data set for teaching machine learning as between 0 – 1. The data is
then imported into Example Set to Tensor to change the data format into the form of vector that can be
processed with Deep Learning (Tensor). The data used for model testing are imported into the Apply
Model to test the performance. They are then imported into Example Set to Tensor in order to change
into Vector form. Then, the data used for teaching the machine and the data for performance testing
are then processed with the Apply Model. Then, the results are imported into the Append in order to
present the accuracy. Within Deep Learning (Tensor), the model connection is defined according to the
RNN technique:

129
 
Deep Learning

Figure 32. Sample model of RNN deep learning

Add LSTM Layer model is implemented so that the model can take the results of previous data to
process along with the current data. It can also remember the processing results which are distant from
the current data, so that the model can predict the results by the data relationship sequence precisely.
Then, Add Fully-Connected is used to compile the data into features that can be processed with the
artificial neural networks. The result of data processing is presented as follows.
In the figure, data scientists can forecast oil prices by using Time Series data to analyze correlation
from previous data. LSTM technique is used to co-process the distanced data to increase the efficiency.
However, the data analyzed with the sample models show accuracy in Root Mean Squared Error form
at 0.546 as shown below.
As presented in the figure, the Root Mean Squared Error value is 0.546. It is highly probable that the
model has predicted incorrectly due to the use of only 10,000 instances for machine learning. Therefore,
when the scientists receive problems that require processing with RNN to find data relationships and
sequences, having sufficient data is another condition that needs considering.

130

Deep Learning

Figure 33. Gas station dataset within RapidMiner software

Figure 34. Sample model of RNN deep learning

131
 
Deep Learning

Figure 35. RNN connection

Figure 36. Results from CNN deep learning processing

Figure 37. Root mean squared error values from oil price forecast

132

Deep Learning

CASE STUDY

Risk Assessment of International Business Organization

Operation Using Deep Learning Techniques

Any business organization usually encounters risks due to uncontrollable external factors such as currency
fluctuations or the threat of war that prevents sellers from producing inputs to deliver. In countries that
need foreign investors to spend in their businesses and make profits until they can return the capital to
investors, the risk assessment is an important issue to guarantee the return to the investors. Internationally
operating companies are all investing and soliciting foreign investors to invest in different ventures. Risk
assessments are therefore necessary for those companies. In 2020, a research called Enabling Legal Risk
Management Model for International Corporation with Deep Learning and Self Data Mining (Wang).
& Chen, 2022) has used the Deep Learning Technique to classify data on legal-related risk issues.
The research adopted deep learning data mining techniques including Gradient Lifting Regression Tree,
BP Neural Network and Support Vector Machine for experiments Performing.
The research has set the finance failure or the opration failure as the target lebel. The results were
divided into 4 levels, namely Giant early warning, Major early warning, Medium early warning and Light
early warning. From the experiment, it was found that the BP Neural Network had good performance
as it identifies the similar level of risk (Actual output) to the determined risk level (Expected output).
Thus, this case study is a good example for data scientists in business and financial institutions who deal
with the huge amount of data on a daily basis. The risk is one factor organizations are all aware. And
data scientists, themselves, either financial experts or non-financial experts, need to study risk assess-
ment frameworks to identify features that may be useful in their assessment in cases the organization
has never collected data useful for risk analysis. other case studies, data scientists still need to rely on
Domain Expert to locate the dataset, and experiment with a variety of algorithms that are most effective
in data analysis.

Acquiring New Customers for E-Commerce Using Deep Learning Technique

In China, which has a vast territory and a large number of cultivation areas in line with the growth of
e-commerce, agricultural products are produced. This has created economic value through selling the
agricultural products to consumers all over China and worldwide through the e-Commerce Platform.
Bringing e-Commerce to help sell agricultural products also allows farmers to cross through middlemen
as they can sell products directly to manufacturers. Each time the consumer receives the product that is
less expensive and the farmer earns more profit. However, in order to increase the value of agricultural
products sales, marketing and finding factors affecting customer purchase through e-Commerce Platform,
has become a challenge for agriculture in China. In 2022, a research called E-Commerce Marketing
Optimization of Agricultural Products Based on Deep Learning and Data Mining (Yang, Zheng &
Sun, 2022) compiled datasets using the questionnaire tool to collect data from 10 respondents who
are farmers in rural China and sell their products through the e-Commerce Platform. The multi-stage
sampling method with 95% confidence was selected. Once completeing the data collection process, the
study received a return of 293 questionnaires, which means this study had datasets of 293 instances.

133
 
Deep Learning

From data collection. it was found that the percentage of respondents selling agricultural products
in wholesale and retail was at 55.7%, wholesale only at 23.5% and retail only at 16.7%, when sold on
the e-Commerce Platform. The problem on selling products are high-priced products at 26.3%, low-
quality products at 65.8%, low customer confidence in the brand at 78.4%, lack of variety at 36.4%, and
inefficient after-sales service at 53.7%. The dataset was then split into a Train Set and a Test Set for
experimental use. In this research, data mining techniques used in experiments are Logistic regression,
Decision tree, GBDT, Neural networks. From such research, it was found that the utilization of Big Data
will enhance the User Experience as the e-Commerce Platform can present diverse product information.
And it ensures buyers, even if it is from a seller who sells agricultural products wholesale. It also allows
buyers to identify the brand of the product based on the user experience. The buyers tend to buy from
the same seller whom they have had experience in the purchase before. It is called “brand loyalty” to
the products. This will gradually increase the opportunity to sell more products. Big Data also extends
to sending marketing messages to customers with a history of purchase orders.
From the agricultural product purchasing behavior through the e-Commerce Platform, where all data
is collected into Big Data, sellers can also identify whether general product visitors have the opportunity
to change to a buyer.
From the case study, it can be seen that the research is not in the nature of collecting data scattered
within the organization or from the Social Network, but the questionnaire tools are carefully designed to
collect data without data collection bias. Collecting data in this manner helps data scientists reduce the
time to do data pre-processing. It is also designed to provide data that can be used to answer questions
honestly. Therefore, for data scientists who analyze data similar to such research, the development of
questionnaires is an alternative to data collection. In cases the organization does not have data that reflects
customer behavior or the existing data need to go through the Data Pre-Processing process until loos-
ing its original pattern, data scientists can design a similar collection of data to overcome that problem.

Using Deep Learning to Forecast Customer Demand in the

Apparel and Fashion Industry (Giri, & Chen, 2022)

Today, customers can browse through the website and place an order so that they can be delivered to
their homes quickly. However, the biggest challenge for apparel business owners is the rapid change in
the needs and tastes of customers. Mass production is at risk of being left in the warehouse because they
cannot be sold due to the very short life of each fashion. In the apparel business, it is necessary to find a
method for calculating the cost of production that is suitable for the lifespan of fashion. Efforts have been
made to forecast such costs and establish pricing to convince the customers to buy in order to drain the
manufactured goods. In the apparel business, products have become so widespread that customers wear
similar clothes.Then the price will decrease accordingly to allow the remaining inventory to be discharged.
Traditionally, the clothing industry has used customer order history to identify the styles of cloth-
ing in each season. However, datasets related to the order history fail to classify patterns of things that
never exist, consistent with the fact that new fashions are unprecedented in the world. Therefore analysis
of such data was not possible to reflect the needs of customers that will occur in the future. What the
clothing industry can do is to bring in other data, data from Social Media to see trends in popularity of
customers as well as other features such as colour, size, and shape, which go far beyond the analysis of
customer purchase history.

134

Deep Learning

In 2020, Chandadevi Giri and Yan Chen (Giri, & Chen, 2022) developed a model to analyze
the fashion needs of customers using deep learning so that the model can classify the data. Classifica-
tion is based on massive amounts of clothing images from past orders and social media in the context
of European fashion.
The researchers have divided the dataset into two groups. The first was numerical data from customer
order history such as order quantity and order frequency. The second group was the images of the clothes
consisted of shirts, skirts and sweaters. The clustering technique is then used to classify the data from
the customer’s order history, and combine the image feature to the result from data defined as the target
variable. In other words, machine learning determines how each fashion image corresponds to a group of
orders. To indicate the consistency, the researchers used the k-NN technique to analyze to which group
of order histories the input was close.
The dataset was then taught to machine learning. In the process of testing the model’s performance,
the researchers inserted newly designed fashion images into the model to test how effective the model
was able to classify groups of orders. After the experiment, the researchers suggest that, in the practical
case, the fashion industry should have a large amount of data to train the machines to have the opportunity
to find a large amount of data enable the machine to classify the data more effectively.
What’s interesting about the case study is that researchers are not only using current data to develop
and test models, but also applying data that machine learning has previously encountered. The experi-
ment on whether machine learning is going to classify the data correctly has developed the model for
real industry use. Deep learning is also applied in a variety of industries and young data scientists should
not ignore the small problems that can be used to solve problems to benefit each industry as well.

Using Deep Learning to Speed Up and Collect Massive Data from loT Devices
Used for Remote Diagnosis of Alzheimer’s Disease (Hannah et al., 2022)

In medical industry, AI, especially image data such as X-ray film, biopsy photographs has been used to
diagnose diseases. Especially in the COVID-19 situation, telehealth counseling has become increasingly
popular. Patients can use devices such as tablets to seek medical advice from doctors. Such services can
also reduce the spread of the disease and infection from going to the clinic or hospital.
Alzheimer’s disease is one of the diseases that affect the elderly and can also be diagnosed through
telemetry. In this case, the medical industry can also use artificial intelligence to obtain patient-related
data to diagnose whether or not a patient has Alzheimer’s disease or stage. The transmission of such
data to the digital medical system has resulted in the effectiveness of artificial intelligence in diagnosing
Alzheimer’s disease. In 2020, researchers have proposed a conceptual framework for optimizing remote
diagnosis of Alzheimer’s disease by applying deep learning to support the speeding up of large-scale
data transport (Hannah et. al., 2022). Data transmission through the block chain is to secure patient data,
which should be confidential as it is personal data that must be transmitted to the digital medical system.
The experiment has used data collected from the device, IoT deployed at patient homes, transmitted
through a block chain to digital medical systems. The tests aims for measuring complete data, transport
efficiency, transmission delay and energy efficiency across a variety of methods.
Therefore, it is obvious that deep learning can also be applied for other types of data such as in the
medical industry where deep learning is used to support the efficiency of transporting large amounts of
data over long distances. And it is also applied in a variety of industries as well.

135
 
Deep Learning

CONCLUSION

Artificial Neural Network data mining technique is efficient for Data Classification. Data scientists can
use features within a dataset to classify data. The features are assigned as a label by machine learning
functionality. Therefore, data scientists can use existing data sets to teach machine learning to create an
artificial neural network. The results are then generated from the data sets that are used to test the model
in order to obtain accurate results. The scientists can also test the accuracy each time, and also apply
the artificial neural networks to classify data in other datasets, such as analyzing borrowers with bank
refinance capabilities, based on the datasets of debtors who have ever borrowed from the banks. There
are two types of deep learning techniques: CNN and RNN. CNN is suitable for analyzing problems that
require data classification. This book puts forward an experiment to classify images starting from num-
bers 0 to 9 using tools within the RapidMiner software including Convolutional Operator, Pooling and
Fully to classify data along with an artificial neural network. For the experiment with RNN, this paper
uses a sample model within RapidMiner software to predict oil prices from time series data sets in order
to analyze correlations and sequences of the data. From the experiments of both CNN and RNN, it has
been found that the amount of data used to teach the machine learning affects the efficiency, accuracy of
data classification and prediction of learning techniques. In addition, processing a large amount of data
also requires GPU Processing to reduce time. When data scientists need to use deep learning techniques
to analyze problems, on-premises digital resources are one thing the scientists need to prepare for the
implementation of data mining techniques used for Big Data processing.

EXERCISE QUESTIONS

1. Download the sample data sets from the Internet for the purpose of classification.
2. From No. 1, do Feature Selection, then manipulate the Missing Value and Outlier.
3. From No. 2, use the artificial neural network model to classify the data and show the model ac-
curacy in the analysis.
4. Define the problems of data analysis that requires processing with CNN
5. From No. 4, find a suitable data set for data analysis.
6. From No. 5, develop an analysis model using RapidMiner software.
7. From No. 6, summarize the results of the experiment and the accuracy of the analysis.
8. Define problems for data analysis that requires processing with RNN.
9. From No. 8, find a suitable data set for data analysis.
10. From No. 9, develop an analysis model using RapidMiner software.
11. From No.10, summarize the results of the experiment and the accuracy of the analysis.

REFERENCES

Aggarwal, C. C. (2018). Neural networks and deep learning: a textbook. Springer. doi:10.1007/978-3-
319-94463-0

136

Deep Learning

Beohar, D., & Rasool, A. (2021). Handwritten Digit Recognition of MNIST dataset using Deep Learn-
ing state-of-the-art Artificial Neural Network (ANN) and Convolutional Neural Network (CNN). 2021
International Conference on Emerging Smart Computing and Informatics (ESCI), Emerging Smart
Computing and Informatics (ESCI), 2021 International Conference On, 542–548.
Chinnarasri, C., Nonsawang, S., & Supharatid, S. (2012). Application of Artificial Neural Networks for
River Stage Forecasting in Hatyai. King Mongkut’s University of Technology Thonburi. KMUTT.
Chow, T. W. S., & Cho, S.-Y. (2007). Neural networks and computing: learning algorithms and applica-
tions. Imperial College Press. doi:10.1142/p487
Fernando, T., Gammulle, H., Denman, S., Sridharan, S., & Fookes, C. (2021). Deep Learning for Medical
Anomaly Detection - A Survey. ACM Computing Surveys, 54(7), 1–37. doi:10.1145/3464423
Ghorpade, V. G., & Koneru, V. S. (2021). Pattern recognition neural networkmodel for experimental
based compressive strength graded self compacting concrete. Materials Today: Proceedings, 43(Part
2), 795–799. doi:10.1016/j.matpr.2020.06.175
Giri, C., & Chen, Y. (2022). Deep Learning for Demand Forecasting in the Fashion and Apparel Retail
Industry. Forecasting, 4(2), 565–581. doi:10.3390/forecast4020031
Guo, C., Zhang, M., & Chen, H. (2021). Suitability of low-field nuclear magnetic resonance (LF-NMR)
combining with back propagation artificial neural network (BP-ANN) to predict printability of polysac-
charide hydrogels 3D printing. International Journal of Food Science & Technology, 56(5), 2264–2272.
doi:10.1111/ijfs.14844
Hannah, S., Deepa, A. J., Chooralil, V. S., BrillySangeetha, S., Yuvaraj, N., Arshath Raja, R., Suresh,
C., Vignesh, R., YasirAbdullahR, Srihari, K., & Alene, A. (2022). Blockchain-Based Deep Learn-
ing to Process IoT Data Acquisition in Cognitive Data. BioMed Research International, 2022, 1–7.
doi:10.1155/2022/5038851 PMID:35187166
Lakshmi Devi, R., & Samundeeswari, V., V. (2021). Detection and Automated Classification of Brain
Tumor Types in MRI Images using Convolutional Neural Network with Grid Search Optimization.
2021 Fifth International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC),
I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC), 2021 Fifth International Conference
On, 1280–1284.
LeCun, Y., Cortes, C., & Burges, J. C. C. (2021). The MNIST Database of Handwritten Digits. Retrieved
from: http://yann.lecun.com/exdb/mnist/
Ramirez-Hernandez, J., Juarez-Sandoval, O.-U., Hernandez-Gonzalez, L., Hernandez-Ramirez, A., &
Olivares-Dominguez, R.-S. (2020). Voltage Control Based on a Back-Propagation Artificial Neural Network
Algorithm. 2020 IEEE International Autumn Meeting on Power, Electronics and Computing (ROPEC),
Power, Electronics and Computing (ROPEC), 2020 IEEE International Autumn Meeting On, 4, 1–6.
Sabau Popa, C. D., Popa, D. N., Bogdan, V., & Simut, R. (2021). Composite Financial Performance
Index Prediction — A Neural Networks Approach. Journal of Business Economics and Management,
22(2), 277–296. doi:10.3846/jbem.2021.14000

137
 
Deep Learning

Snineh, S. M., Amrani, N. E. A., Youssfi, M., Bouattane, O., & Daaif, A. (2021). Detection of traffic
anomaly in highways by using recurrent neural network. 2021 Fifth International Conference On Intel-
ligent Computing in Data Sciences (ICDS), Intelligent Computing in Data Sciences (ICDS), 2021 Fifth
International Conference On, 1–6.
Sriwiboon, N. (2020). Improvement the Performance of the Chest X-ray Image Classification with Con-
volutional Neural Network Model by Using Image Augmentations Technique for COVID-19 Diagnosis.
The Journal of King Mongkut’s University of Technology North Bangkok, 31(1).
Telikani, A., Shahbahrami, A., & Gandomi, A. H. (2021). High-performance implementation of evo-
lutionary privacy-preserving algorithm for big data using GPU platform. Information Sciences, 579,
251–265. doi:10.1016/j.ins.2021.08.006
Thankachan, T., Prakash, K. S., & Jothi, S. (2021). Artificial neural network modeling to evaluate
and predict the mechanical strength of duplex stainless steel during casting. Sadhana, 46(4), 1–12.
doi:10.100712046-021-01742-w
Wang, G., & Chen, Y. (2022). Enabling Legal Risk Management Model for International Corporation
with Deep Learning and Self Data Mining. Computational Intelligence and Neuroscience, 2022, 6385404.
doi:10.1155/2022/6385404 PMID:35432517
Yang, H., Zheng, Z., & Sun, C. (2022). E-Commerce Marketing Optimization of Agricultural Products
Based on Deep Learning and Data Mining. Computational Intelligence and Neuroscience, 2022, 2022.
doi:10.1155/2022/6564014 PMID:35634060
Zhu, H., Samtani, S., Brown, R. A., & Chen, H. (2021). A Deep Learning Approach for Recognizing
Activity of Daily Living (Adl) for Senior Care: Exploiting Interaction Dependency and Temporal Pat-
terns. Management Information Systems Quarterly, 45(2), 859–895. doi:10.25300/MISQ/2021/15574

138
 139

Chapter 7
Clustering

ABSTRACT
Clustering is employed to divide a data set into an appropriate number of groups. Clustering is a form
of unsupervised learning, which means a data scientist can bring labelled features of interest into the
mining model. Furthermore, after dividing the data set, the data scientist can label each cluster. In busi-
ness, clustering is used to analyze a customer or product segment that matches a target market. This
chapter introduces clustering techniques including k-means, hierarchical clustering, and DBSCAN as
well as techniques to indicate the efficiency of the clustering analysis. Data scientists can assess the
efficiency of clustering analysis in two ways. Firstly, subjective measurement is where a data scientist
consults a domain expert to confirm the efficiency of the cluster analysis, and secondly, data scientists
can use objective measurements that test the efficiency of the cluster analysis result based on calcula-
tions. This chapter demonstrates cluster analysis adoption with RapidMiner so that readers can follow
the process step-by-step.

INTRODUCTION

Clustering is Unsupervised-Learning data mining technique (Govindaraj et al., 2020: Abbas et al, 2021).
Data scientists are able to group the quantitative data without defining the target variables (Labeled) nor
dividing the dataset into 2 parts to teach the machine learning. Clustering technique gather similar data
into one group and bring dissimilar data into another group. Therefore, clustering can be applied in a
wide range of industries, such as customer segmentation according to customer purchasing behaviors
(Punhani et al., 2021). The customers in the same group tend to have similar purchasing behaviors and
expected prices. Clustering is applied in order that the business sectors can design marketing plans suit-
able for each group of customers. It can also diagnose the stage of cancer in each patient from the sizes
of the tumors growing in the patient’s organs (Kumar, Ganapathy & Kang, 2021).
Although the process of data science provides computational methods to obtain the optimal number
of clusters, to specify the number of clusters depends on the objectives of the analysis. Data scientists
need to consult with domain experts to gain the exact number of clusters to be analyzed. For example,
clustering the sizes of shirts that will be produced by analyzing the clustering data from the purchase

DOI: 10.4018/978-1-6684-4730-7.ch007

Copyright © 2023, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
 
Clustering

history datasets. In this way, data experts can advise on a number of clusters of shirt sizes that are ap-
propriate, such as small, medium, and large.
Clustering can be used to analyze the centroid of the data in each cluster. It can be applied to set the
car sales price from the purchase history. Each car model can be produced in many specifications. After
the analysis, the car dealers can set various prices of each specification under the same car model by
using the data in each group. For example, Model A car with the entry-level specification is 1,300,000
baht. Model A car with the intermediate specification is 1,600,000 baht. Model A car with the high-
est specification is 1,700,000 baht. As mentioned in the example, the price determination is from the
analysis using clustering technique. Data scientists can perform a centroid analysis to each group of data
according to the customer purchasing behavior.

K-MEANS CLUSTERING

Once data scientists are asked about the nature of the data, such as the purchasing behavior of each group
of customers, they need to find a consistent data set to answer those questions (Ginting, 2021: Puspasari
et al., 2021). The dataset initially received is characterized as Unlabeled Data, the data which has not
yet been clustered nor defined for its name or meaning. Therefore, these data sets can be clustered and
defined according to the objectives of the data analysis, such as definition of the customers; the middle-
class or the wealthy.

K-Means Clustering Algorithm

Clustering data can be done by bringing 2 Data Objects; the shirt width and the shirt length; to create
a Data Point; the sizes. Data scientists can develop a display in the Scatter Plot format to see the inter-
section at the data point created from the 2 data objects, and then colorize each data point according
to clusters. For example, the yellow data point represents the medium size, and the brown data point
represents the large size.

Figure 1. Unlabeled data and data after clustering

140

Clustering

To cluster data using the K-Means algorithm, data scientists can import the quantitative datasets
into the K-Means model, then determine the number of clusters, and perform clustering analysis. The
process is as follows.

1. Determine the number of clusters into k variables. For example, k = 2.

2. K-Means model randomly places 2 zero points into the data set, assuming that the 1st center is C0,
and the 2nd center is C1.

Figure 2. Clustering data with k-means

3. K-Means model will then calculate the distances from each data point to the center. For example,
it will be calculating the distance between center C0 and data point a, and calculating the distance
between center C1 and data point a. Meanwhile, K-Means calculates the distance between center C0
and data point b, and calculates the distance between center C1 and data point b. When data point
a is closer to C0 than C1, data point a is then clustered in Cluster 0. Similarly, when the distance
between data point b is closer to C1 than C0, then data point b will be clustered in Cluster 1.
4. When the data points are grouped in the cluster form, K-Means will then calculate the average
distance from the center point of each cluster to all data points, and then move the position of the
two centers again. Then, it will perform the calculation in step 3 and compare the average distance
from the center to the data points of each cluster from the average of the previous calculations. If the
mean varies or greater than 0, the clustering has not yet been completed. So, K-Means will repeat
the process over and over so that the data point is positioned closest to the center of each cluster.
In case the average is equal to the previous round, or equal to 0, clustering process is considered
completed, and K-Means will no longer relocate the center point.

141
 
Clustering

CALCULATING DISTANCES BETWEEN THE DATA

POINTS WITHIN K-MEANS CLUSTERING

The distances between the data points and the center of each group within a K-Means cluster can be
calculated in two ways: Manhattan Distance and Euclidean Distance.

1. Manhattan Distance: Is a method for calculating distances in quantitative data by subtracting the
data objects from each instance, and adding the results of all data objects together as presented in
the example (Haviluddin et al., 2020).

Table 1. Examples of datasets for Manhattan Distance calculation

Name Age Bank Balance Credit Rating

Jirapon Sunkpho 50 550 3
Sarawut Ramjan 38 520 2

From the Manhattan Distance table, the Data Object from Instance named Jirapon Sunkpho will
be diminished by the Instance named Sarawut Ramjan, starting from the Data Object named Age. 50
decreased by 38 equals 12. The Bank Balance Data Object will then be subtracted. The Subtraction is
550 diminished by 520 equals 30, and the Data Object from Credit Rating is subtracted, being 3-2=1.
Then all of the calculated results are added as 12 + 30 + 1 equals 43, which is the distance between the
instances named Jirapon Sunkpho and Sarawut Ramjan.
The Manhattan Distance method is, therefore, used in data clustering techniques to group the pur-
chasing behavior or to classify data to segment products. The Manhattan Distance method calculates
the distances between data objects in four directions: left, right, top and bottom.

2. Euclidean Distance: Within the data set, there is movement of data objects while clustering data
in multiple directions. Therefore, data scientists also need to use Euclidean Distance to calculate
distances between data objects in various (Bayram & Nabiyev, 2020).

Table 2. Examples of datasets for Euclidean Distance calculation

Name Age Bank Balance Credit Rating

Jirapon Sunkpho 50 550 3
Sarawut Ramjan 38 520 2

From the sample data set, to calculate the distance between the instances named Jirapon Sunkpho
and Sarwut Ramjan, the data object of each instance can be calculated as follows:

142

Clustering

dis (x , y ) = (50 − 25) (550 − 520) (3 − 2)

2 2 2

dis (x , y ) =39.06

The distance between the two instances is 39.06. In the case the data objects are literal, and data sci-
entists can replace the identical data with the number 1, and the different data with the number 0 before
calculating with the usual Euclidean Distance method. This is called Mix Euclidean Measure Distance.
Regarding segmentation using K-Means, data scientists are required to take into account outliers,
which are used to calculate the distance between the data point and the center. Therefore, data scientists
need to prepare data (Data Pre-Processing) before segmenting them.
However, data scientists are supposed to consider the nature of clustering in the data set before de-
termining the appropriate number of clusters. For example, most of the customer’s purchasing behavior
data is clustered into two groups, each group having a high-density data point. If data scientists define
three groups, the third group may contain only a small number of data points that are insignificant to
use as they do not reflect the results that support decision-making.

HIERARCHICAL CLUSTERING

Hierarchical Clustering is a principle that supports the decision making of data scientists (Rezaeijo,
2021: Zhou et al., 2021). To determine the number of data clusters, there are 4 methods for calculating
the distance between the data clusters; 1. Calculating the distance from the Data Point in each group
that is the closest to each other, 2. Calculating the distance from the Data Point in each group with the
farthest distance, 3. Calculating the distance from the average distance of each group, and 4. Calculating
the distance between the centers of each group.
Agglomerative Clustering Algorithm is a clustering method that supports Hierarchical Clustering.

1. Bring Data Object with 2 or more variables to create Data Points which can be displayed through
the Scatter Plot in order to see the location of the created Data Points such as the number of cus-
tomers and food sales. Then the customers can be clustered.

Table 3. Sample data for agglomerative clustering algorithm (Munlika, 2019)

Number of Customers per Day (X) Food Sales (Y)

120 100
59 50
86 70
35 20
105 90

143
 
Clustering

From the table, the data scientists can develop the visualization data to explore the data on the dis-
tances between the data points as shown below.

Figure 3. Creating a scatter plot to explore shopping trends

2. Clustering analysis can be done by choosing a method to calculate the distance from the nearest
Data Points in each round to cluster the data with the same distances in each group. It can be cre-
ated as a Proximity Matrix showing the distance between the Data Points as well as the number of
clusters as follows:

Figure 4. Proximity matrix, the result of calculating the distance of every pair of data points

As seen in the figure, to calculate the distance between each data point within the Proximity Matrix,
data scientists can divide the data into clusters. In the example, Cluster 1 is Instances 1 and 5, which
share the same distances between data points, 18.02. And Cluster 2 is Instances 2 and 3, which are the
distances between the data point, 33.60. The third cluster is Instance 4, which covers the 2nd and 3rd

144

Clustering

instances. Therefore, in the example, there are 4 clusters, which segment the data of the number of cus-
tomers per day together with food sales. When considering the data, it is found that the original data,
grouped into each cluster, shares similar volumes. The graph below presents the segmentation as follows.

Figure 5. Clustering with hierarchical clustering

In the figure, the data are clustered by calculating the nearest distance to the nearest of data point,
clustering from the nearest data point first and then to the distance that covers all clusters (Peungvicha,
Tubtiang, & Sattayaaphitan, 2019). From the picture, data scientists can decide to cluster the data in
either of these ways.

1. The cluster consists of cluster 1 that covers instances of the 1st and 5th instances, and cluster 2
covering the 2nd, 3rd and 4th instances.
2. The cluster consists of cluster 1 covering the 1st and 5th instances, cluster 2 that covers the 2nd
instance, and cluster 3 covering the 2nd, 3rd and 4th instances.

However, clustering with Hierarchical Clustering has limitations. If the data scientists are unable to
pre-process the original data sufficiently and the outlier remains, the data is computed for clustering
with other data too. This results in discrepancies in Hierarchical Clustering. Moreover, by calculating
the distances between every data point within the datasets, data scientists need to provide sufficient
resources as Big Data manner in order to be clustered with Hierarchical Clustering method.

DBSCAN

In some cases, when data scientists survey data by creating a scatter plot and have found that the data is
highly distributed. Data scientists can use DBSCAN to cluster their data (Wafa et al, 2021). DBSCAN
will determine the center radius (Eps) from every data point, and determine the minimum number of the
Data Points that must be within the radius (MinPts). Therefore, DBSCAN aims to determine the density
of data within each cluster with the following algorithm.

145
 
Clustering

1. DBSCAN assigns all data points as the center by analyzing the data points that are within the range
of Eps whether the number exceeds MinPts or not. If the data points are greater than the MinPts,
then the central data point is considered the Core Point. For example, if MinPts is 4, DBSCAN
will analyze whether there are more than 4 data points within the radius. And if so, it will assign
the center data point as core points and do the same for all data points within the datasets for de-
termining data cluster.
2. In the case DBSCAN analyzes and have found that within the radius of Eps, the number of data
points is not greater than MinPts, it will set the Data Points as Border Points. The border points
are then clustered within any Core Point to combine into one cluster.
3. If a Data Point is found near the Core Point more than 2 locations, it will cluster those data into
one group.
4. If it is found that within the Eps radius from the central Data Point is fewer numbers of Data Points
than the MinPts, and Data Points at the center does not cling to the Core Points, any Data Point
will be designated as Noise.

Figure 6. Clustering data with DBSCAN

In the figure, the data scientists have set the MinPts value to 4. When the radius within the radius
of Eps meets 4 center data points, then the data points are assigned as the core point. And DBSCAN
will get the data cluster 1. Then, when setting the center within the Eps radius again, only 2 Data Points
can be found from the center. Such center points are designated as Border Points and then grouped into
Cluster 1. When the center is set within the radius of Eps again, only 2 Data Points are found, and the
cluster is not grouped with other Core Points, so the center point is considered Noise. To determine the
Eps or radius from the data point set as the center, data scientists can determine the Eps configuration
based on the distance from the center to other data points within the most frequent dataset. It means
within most datasets, the distance from the center to the data point is equal to that value. For example,
when analyzing the data, it is mostly found that the distance from the center to the data point is 0.8. The
data scientists can assign Eps to 0.8, and take all the distances to create a histogram for analyzing the
greatest distance frequency.

146

Clustering

ANALYZING THE EFFICIENCY OF DATA CLUSTERING

It can be done in two ways, 1. Subjective measurement, which is an assessment from data experts. Experts
may consider the data with their experience along with the results of data visualization, and 2. Objec-
tive Measurement, which is to consider from the numbers that show the efficiency of data clustering.
Each cluster must be clearly grouped, and within each cluster there must be consistent data. To analyze
efficiency, the details are as follows.

1. Subjective Measurement: Is to present visualization data to data professionals to help assess the
efficiency of data clustering. Visual data from data mining techniques can be used for data clas-
sification such as Decision Tree or visualization data resulting from cluster analysis. Data cluster-
ing must be consistent with the context of the industry. For instance, after clustering the data, data
scientists can develop a data visualization that uses color to represent each cluster. Cluster 1 is
represented by red, and Cluster 2 is represented by blue. Also, in each cluster there is a Centroid
which shows the center of the data in each cluster. Data professionals can determine whether data
clustering is accurate, reasonable and consistent with the business context. For example, data clus-
tering of customer pants sizes is divided into 3 clusters, consisting of sizes S cluster, L and XL. It
is considered accurate if each cluster has clear and reasonable data on width and length of pants.
Therefore, the customers can purchase pants using their individual size.
2. Objective Measurement: Uses numbers to analyze the performance of three types of clustering;
Dunn Index, Davies-Bouldin Index and C-Index.
a. Dunn Index determines the minimum distances of the Data Point to the Data Point in other clus-
ters (Intercluster) divided by the maximum distances of the Data Points within the Intracluster
(Ben Ncir, Hamza, & Bouaguel, 2021).

dmin
d=
dmax

From the formula, dmin is the minimum distance of the Data Point to the Data Point in other clusters.
dmax is the maximum distance of the data point within the cluster.
The result shows that the larger the number, the better the clustering efficiency. This means that the
distance between the clusters is large, making the clusters are further separated. And when the distance
between the data points is less, the data within the Cluster is well grouped.

b. Davies-Bouldin Index is the analysis of data clustering using data from the centroids of every
cluster. The algorithm is as follows (Wijaya, 2021)
i. Calculate the average distance from the Centroid to all Data Points within the same Cluster.
ii. Calculate the average distance from each centroid of each cluster to the other cluster’s centroids.
iii. Take the average distance from the Centroid to all Data Points of Cluster 1 to add with Cluster
2 and divide by the distance between the Centroids of the 1st Pair.
iv. Continue this algorithm until all pairs are complete, and then select the result from the most
valuable pair. Then, take the above result to divide by the total numbers of clusters.

147
 
Clustering

The results can reflect the efficiency. The lower the values are, the further the clusters can be sepa-
rately grouped. Each pair should have a considerable distance between the Centroids, and within each
cluster there is good clustering of data. And the distance between Centroid and Data Point within the
same cluster is small.

c. Index is an analysis of clustering efficiency, which can avoid discrepancies from data points, the
noise as follows:

S −Smin
c =
Smax − Smin

when, S is the sum of the distances of each data point within the same cluster
Smin is the sum of the smallest distances of each data point from all clusters
Smax is the sum of the largest distances of each data point from all clusters

When calculating the results are between 0 – 1, small results mean good clustering efficiency. If the
sum of the distances between the largest data points shows a high value, it means that they are clustered
separately from each other, and if the lowest distance between the data points is small value, it means
that the data points within each cluster is clustered appropriately.

DATA CLUSTERING USING RAPIDMINER SOFTWARE

To use the K-Means clustering software, RapidMiner in clustering, this book uses the Irish flower data
set to group the Irish flower data of each cultivar including Setosa, Vesicolor and all Virginega, total-
ing150 instances. K-Means is classified as Unsupervised-Learning where data scientists can import the
datasets into the model without labelling the target variables nor teaching the machine. The experimental
process is as follows.

Figure 7. Importing the Irish flower dataset into the model

148

Clustering

In the figure, data scientists can import the Irish flower dataset within the RapidMiner software into
the model for experimentation. The Irish flower dataset contains the following information.

Figure 8. Irish flower data set

Figure 9. Using k-means clustering

149
 
Clustering

In the figure, data scientists use the K-Means Clustering model to cluster the data by setting the
number of clusters to 3. The distance between the data points can be calculated using the Euclidean
distance method. The results of the initial clustering are as follows.

Figure 10. The number of data objects organized in each cluster

From the figure, the data sets are divided into 3 clusters as defined. Cluster 0 has a total of 50 Data
Objects, Cluster 1 has a total of 38 Data Objects, and Cluster 2 has a total of 62 Data Objects. In each
cluster, there are clustered data as seen through visual information below.

Figure 11. Data results after k-means clustering

150

Clustering

As seen, data scientists can take the visualization data to consult with a data specialist in order to
assess the effectiveness of the data clustering, considering the distance between clusters, obviously
separated and clustered within the cluster. The Irish flower cultivars are clearly divided into 3. And the
sizes of flowers and sepals in each species totally differ.
However, to reconfirm the efficiency of data clustering, data scientists can also perform additional
analysis.

Figure 12. Measuring clustering efficiency with cluster distance performance

In the figure, data scientists apply Cluster Distance Performance to assess the efficiency of data
clustering. The average distance between the data objects within the cluster is evaluated. The average
distance between the Centroids in each cluster, and the overall performance are as follows.

Figure 13. Evaluation of clustering efficiency result

In the figure, when using Cluster Distance Performance model to assess the efficiency of clustering,
the following results are obtained. The average distance between the centroids in each cluster is 0.52.
The average distance between data objects within cluster 0 is 0 0.30. The average distance between data
objects within cluster 1 equals 0.62, while the distance between data objects within cluster 2 is 0.64.
Last but not least, the efficiency calculated by Davies Bouldin method is 0.66.
From all the results, each pair of clusters have good distances between the Centroids, meaning the
clusters are appropriately separated from each other. And within each cluster, the distance between

151
 
Clustering

Centroid and Data Object is not too high, which means there is a moderately grouping effect. When cal-
culating with Davies Bouldin method, the results reflect the efficiency of clustering at a moderate level.

CASE STUDY

The Smart Travel Route Design With Clustering Analysis Algorithm (Lou, 2022)

China has achieved a lot of success in the tourism industry since its opening. However, in 2022, tourism
in China is beginning to reach its saturation due to the lack of analysis of tourist routes. The analysis of
tourist routes is the planning of visits to various attractions to meet the specific expectations of tourists,
such as health tourism. The attractions along the route are places related to the health of tourists, such
as natural hot springs and therapeutic spas, which are approved and maintained by the tourism indus-
try related government agencies. To design of the tourist route, it is important to take into account the
circumstances surrounding each period; rainstorms or earthquakes that may occur. It then produces an
action plan that tour operators can implement.
In 2022, a study called Analysis of the Intelligent Tourism Route Planning Scheme based on the
Cluster Analysis Algorithm (Lou, 2022) conducted a study on the analysis of tourist routes by dividing
the tourist routes into 3 scales, namely large, medium and small scales. For the small scale, tour opera-
tors need to analyze the needs of tourists in each form, aiming to provide tourists with whether it’s a city
tour or the historical and cultural tour in each locality. Tour operators and travel agents can consider the
tourist types by their interests. The tourists’ preferences can be categorized as shopping, recreational
tourism, adventure tourism and historical and cultural studies.
To analyze such tourist routes, there is a great challenge of providing tourist destinations that meet
the expectations and various tastes of tourists to come together on the same route. Therefore, there has
been researches aimed at using Big Data for in-depth data analysis. However, if data scientists get travel
datasets from a time-consuming travel together with a not-so-good experience to tourists, they will not
be able to answer the question. Such datasets simply cannot provide insight to data scientists because
data scientists themselves cannot expect how complete the datasets they receive will be. Therefore,
there is a number of researches trying to suggest improving the algorithms for data processing so that
the incomplete data can be used to process the data by using the ant colony algorithm together with
the travel route design. There are also a number of studies that have used a particle swarm algorithm to
design a travel route with the shortest time frame by choosing the attractions that are close to each other
so that tourists can experience travel at the lowest travel costs. Therefore, designing the route in such a
way will allow tourists to travel qualitatively with a small amount of time and save the cost of the tour
operator. However, the variables that affect the traveler’s experience are not only attractions and dura-
tion of travel, but the age range of tourists between teenagers and the elderly in the same group is also
a challenge for tour operators and travel agents to design travel routes as well. A research analysis was
conducted on the tourist personalization route guidance system.
Regarding the condition of China’s tourism industry today, although the design of tourism routes is
a principle and standard, accepted all over the world, in the real situation in China, most tour operators
and travel agents do not apply those principles to the design of tourist routes because each locality has
so many variables and needs that those principles cannot be applied for route design. Therefore, tour
operators and travel agents use the experience of their staff as a resource to design travel route. Therefore,

152

Clustering

tourist routes often face congested traffic conditions because Staff are just familiar with those routes on
the journey. On the other hand, China’s attractions focus on recreational tourism and sightseeing. As a
result, when designing tour routes, tour operators and travel agents do not have many options to meet
the different expectations of each group of tourists. Designing a good tourist route should also take into
account the delivery of experiences about Food, accommodation, cultural values, the convenience, and
the distance between the tourist’s accommodation and the attractions.
Cluster analysis techniques are, therefore, used for the design of tourist routes that not only manage the
tourist attractions under the routes and timeframes, but also bring cultural values that meet the expecta-
tions of tourists. Clustering focuses on segmenting data. The data in each cluster can be distinguished
from the properties used in the data analysis. Clustering in data science is similar to processing with other
data mining techniques. It begins with selecting datasets that are useful for answering the set questions
and attribute transformation for a selected data set to become a feature used for processing with data
mining techniques. Then only the features that are useful for data analysis are selected. This may include
performing data cleaning, such as eliminating outlier data objects or adjusting the measurements of each
feature to be consistent, called the standardization. The data is then analyzed with Cluster analysis, which
is Data Mining in an Unsupervised-Learning manner, and then interpret the results.
The research selected techniques corresponding to experiment cluster analysis in 4 methods: density-
based cluster analysis algorithm, division-based cluster analysis algorithm, hierarchical cluster analysis
algorithm, and k-means cluster analysis. It was found that using k-means cluster analysis is easier to
interpret the data obtained from the analysis. It takes less time to process the dataset and low complexity.
However, with k-means cluster analysis, outlier data still affects the efficiency of processing and interpret-
ing the results. The research has, therefore, improved the algorithm for further processing by identifying
a group (class) of the city as a tourist destination and calculating the distance between those classes,
together with calculating the distance of the Data Object related to expectations and personalization of
tourists. The key principle is that the smaller the distance, the more data objects are grouped together.
The distance between both the attractions and the tourist-related data objects were processed together
by k-means clustering until it is possible to cluster the attractions suitable for each type of tourist. By
improving the algorithm in this process, tour operators and travel agents are able to use the analyzed
attractions to create a travel itinerary that meets the tourists’ expectation.
Therefore, besides clustering in the various ways that this book has already mentioned, Data scientists
can also make improvements to processing algorithms to make Clustering Analysis flexible and suitable
for different objectives and environments.

Using Cluster Analysis to Segment Customers Based

on the Nature of Engagement Created by Social Media
Influencers on YouTube (Erdoğmuş & Arslan, 2022)

Social media creates dynamic relationships between brands and customers, so some customers can use
that relationship to influence the perceptions of other customers, which we call influencers. Influencers
can spread their opinions about the products and services they’ve tried or reach their followers. Ad-
vertisers and marketers believe in the influencers’ ability to communicate about a brand to their target
customers. However, choosing the right influencer for customers and brands has also become a chal-
lenge for marketers. The metrics used for choosing an influencer on a social network are the number of
Like, Share, and Subscriber as a measure of the effectiveness of each influencer’s brand communica-

153
 
Clustering

tion. Understanding the influencer’s target audience will help influencers design the right content and
marketing for their followers.
These issues become a challenge for brand managers to understand the phenomenon of social media
and how to take advantages from the engagement that influencers generate to their followers. For the
influencers, understanding their target audience can help influencers design the right content and mar-
keting for their followers. YouTube is one of the most popular social media platforms worldwide and
is used to publish content and engage your influencer’s customers to help them make the purchase of
products and services reviewed by the influencer.
The idea of ​​engagement is to create a two-way relationship between the brand and the consumer. That
means the consumer influences how the product and service preferences meet their expectations, and
the brand expects their customers to have a positive attitude towards their own brand, which is reflected
through the use of the brand’s products and services. When it comes to online engagement, the brand
can review the behavior and attitude that customers have towards the brand through review, like, dis-
like, follow, share, comment, subscribe, participate in online games. The brand manager can categorize
customers on social media based on the nature of their engagement into 6 categories: those who do not
use social media, those who travel and watch media on social media, those who interact with influencers
on social media, those who surf on social media surfers to collect information on purchasing decisions,
those who use social media to review products and services via social media, and finally those who
create social media contents.
When it comes to social media motivations, the users benefit from choosing to consume content and
use content on social media, interacting with people on social media, searching for content on social
media for educational purposes, exploring content on social media for news or new topics, content, kill-
ing their boredom, consuming content for entertainment, accessing to social media to separate time from
day-to-day operations, creating one’s own content to be accepted by others on social media, entering
the online world to distance themselves from offline society, building relationships with others in the
online community, using social media as a medium to communicate with people in society anywhere
and anytime, and exploiting social media such as presenting themselves to be known in the hope of
marketing results. These uses are all due to the motivation for using social media, and when specifically
mention to YouTube, the main motivation of users is to consume entertainment media, where users may
both upload their own video files to communicate to other users and choose to consume content created
by other users.
In 2022, there was a research called “Engaging with Social Media Influencers on YouTube: A Clus-
ter Analysis” (Erdoğmuş & Arslan, 2022, which attempted to segment customers based on the type of
engagement that YouTube influencers generate using a cluster analysis technique. The research used a
questionnaire as a tool to gather information about YouTube users’ motivation to engage with influenc-
ers. Then K-Mean cluster analysis techniques were used to analyze the data. The results showed that
customers can be divided into 3 groups: 1. Customers with positive behavior: This group of people use
YouTube for entertainment and often don’t press the dislike button on YouTube and don’t pay atten-
tion to the amount of people who press dislike. They also often don’t press the unsubscribe channel on
YouTube and often press Like and Share more than other customers. 2. Passive group: These people
still use YouTube for entertainment. But their behavior is more like content consumption than interact-
ing or communicating with influencers and other users. There is a chance for this group of people to
click unsubscribe channel on YouTube. 3. Users who access YouTube for entertainment and search for

154

Clustering

information about goods and services solely without any interaction: They often consider the quantity
of likes and dislikes in making decisions about purchasing goods and services.
Therefore, influencers and marketers or brand managers should pay attention to creating content and
storytelling for positively behaved customers who are interested in following the influencer in order to
establish their own social status, make other users interested and purchase goods and services based on
the influencer’s recommendations. For customers who seek information about products and services on
YouTube, behaviors caused by positive behaviors of other users such as like or subscriber result in their
decision to purchase goods and services from influencers.
According to the results of the study, data scientists who analyze marketing data can also consider
features that may respond to the analysis of market data. It’s not limited to demographics. Based on
case studies, segmenting customers by engagement level requires a combination of data analysis across
multiple features, and it can be demonstrated that if a data scientist wants to analyze data that is deep
in a particular science, it is important to analyze data that is deep in a particular science. Understand-
ing only quantitative data analysis or advanced data mining techniques cannot enable data scientists
to obtain data of an Unseen or hidden relationship nature. When a data scientist goes into a particular
department or organization, to find knowledge in the domain will help data scientists develop themselves
to be proficient in data analysis.

Using Cluster Analysis to Analyze After-Sales Service (Hyrmet & Arbana, 2021)

Digital transformation requires multiple variables, and among those driving variables is creating positive
customer experiences and creating new business models that meet customer expectations. Organizations
that are aware of digital transformation will focus on tightening internal processes and reducing duplication
of work between departments, and creating an environment within the agency ecosystem that facilitates
innovation, including enhancing the digital maturity of the organization through data governance.
In order to consider digital transformation, organizations need to consider it in a variety of dimen-
sions. 1. In terms of digital strategy, organizations need to take into account the formulation of vision
and data governance and strategy implementation to drive the organization towards the vision. 2. In the
dimension of creating the ability to deliver a positive customer experience, organizations must perform
customer segmentation analysis, using data from social networks to analyze data and improve prod-
ucts and services in line with customer and subscriber expectations. Organizations can use forecasting
techniques to formulate marketing strategies, as well as create flexible customer service processes. 3.
Technological dimensions involve planning the use of technology, integrating the use of technology into
the business model of the organization. 4. Operational dimensions aim to consider the development of
internal processes, the implementation of technology for responding to management strategies, as well as
increasing efficiency and productivity from improved processes. 5. In terms of organizational manage-
ment, the organization must focus on changing the corporate culture to be in line with the organization’s
performance. Digital Transformation communicates such changes throughout the organization, provid-
ing training to develop employees to be ready for the adoption of digital technology for their operations
and managing knowledge to inherit and transfer valuable skills and knowledge within the organization,
independent of employee entry and exit.
As a result of the organization’s digital transformation efforts, after-sales service is another service
that organizations need to consider in order to create a service strategy that affects the satisfaction of their
customers. Therefore, in 2021, Hyrmet & Arbana conducted research using Cluster analysis to segment

155
 
Clustering

customers based on their experiences gained from the after-sales service so that the data can be used to
develop an appropriate an after-sales service strategy.
The research report seeks to study a variety of data mining techniques, Classification or Clustering,
that can be used for segmenting customers based on their experiences provided by the service. It was
found that the Clustering technique is suitable for analysis in response to research. Within the Cluster-
ing Technique, there are many consistent methods, K-Mean Clustering, farthest-first algorithm, Canopy,
clustering algorithm, Density-based group of clustering algorithm, and Filtered clustered Determining.
To consider which algorithm is suitable for that task, the researcher begins with an understanding of the
business by considering the types of businesses that need to provide the after-sales services by retailers,
freight forwarders and factories who sell inputs. Then, the touch points that customers encounter to receive
after-sales services, including call center, freight forwarding, and product repair and maintenance, are
considered. After that, the researcher has searched for the Data Set for the experiment which is consistent
with the research objectives. The Data Set consists of the following Features:

• Product Warranty
• The nature of the maintenance service of the goods is divided into the provision of services pro-
vided by the supplier and the outsource to the partnership.
• Image of products and services
• Characteristics of after-sales service
• Types of products such as furniture, domestic products, imported products
• Product life
• Maintenance and maintenance periods
• The sum of the amount of time for the maintenance, guarantee. The sum of such periods must be
0.

After that, we used Feature Selection techniques to reduce features that were not useful for data analy-
sis. A total of three used methods are Correlation-based attribute evaluations, information gain attribute
evaluations, and wrapper attributes evaluation each of which yielded results by reducing the number of
different features. The remaining features were then processed with cluster analysis techniques across
all four of the methods studied. The farthest-first algorithm showed the highest accuracy, which took a
short time in processing.
Based on the analysis of such data, businesses need to plan for digital transformation in a variety of
dimensions to consider gaps they can use digital technology to improve processes in order to deliver a
positive customer experience. The data mining techniques that are suitable for data analytics are then
selected so that organizations can use the results to determine products and services and deliver value
to customers to meet customer expectations.

Improving the Efficiency of Cinematic Review Data Analysis

With the Use of Cluster Analysis (Xu et al., 2022)

On the site for movie review, viewers can freely press reviews to score a movie without any pressure,
as well as write textual comments to criticize the movie they’ve watched. The review scores and com-
ments have become a variable that makes it possible for other viewers who haven’t watched the movie
to decide whether to watch or not watch it. Marketers are using reviews and comments to make viral

156

Clustering

content. The variables that affect the decision to watch a movie consists of: Cinematic content, the
source of the content related to the film, the genre of the film, and the timing of its release. Analyzing
data from review scores and comments has become an issue for data scientists to understand audience
behavior in order to use the results to promote marketing. Data analytics and data visualization are an-
other way to help marketers understand data patterns. A good movie is often made with a large number
of elements, such as a sensible plot, the actors’ performances, spectacular scenes, and the soundtrack.
Data scientists can extract comments that correspond to keyword-based data analysis, such as stories or
events. Similarity of word groups should be classified into one group. The data pre-processing is a way
for data scientists to extract groups of words that are characterized as big data and then analyze it with
cluster analysis. Data scientists can use Binary Classification, Decision Tree or Naïve bays, to identify
which words are keywords.
In 2022, there was a research study on data analysis for marketing promotion of movies with movie
review data (Xu, Chen, & Yang, 2022). The audience’s satisfaction level corresponds to the real human
emotions that are ambiguous and cannot be classified as right and wrong or black and white. Then, the
keywords within the dataset is extracted; positive comments or negative comments. In accordance with
the application of Fuzzy Logic theory, it can be determined that each keyword is classified in a com-
ment type; the comments that are rather bad, the comments that are objectively based on the facts, the
comments that do not express either positive or negative emotion, Comment in a good way and the com-
ments that go in the way of admiration or delight in the movie Therefore, another prerequisite of such
research is the application of Sentiment Analysis to enable machines to determine whether a keyword
is a positive or a negative comment.
In the process of dividing keyword data with the K-Clustering Technique, researchers or data sci-
entists have also apply Principle Component Analysis: PCA to reduce the number of data dimension to
increase data division efficiency and display data through data visualization.
The research used clustering techniques to break down keyword data in the form of data visualization
so that marketers could understand audience emotions and experiences reflected in the form of comments
and film ratings, and could use that information to further improve film production and marketing pro-
motion. The research also applied both Fuzzy Logic and Sentiment Analysis in the data pre-processing
to increase the data processing efficiency of the clustering technique and attribute Transformation to
make data ready for processing with data mining techniques.

CONCLUSION

Clustering is used to classify quantitative data and applied in various ways, such as customer segmentation
and classification of diseases. Data scientists can import quantitative data into RapidMiner software, and
use data mining techniques such as K-Means Clustering or DBSCAN to cluster the data. In a good data
clustering, the data within each group should be clustered, but each cluster should be clearly separated
from each other. Within RapidMiner software, it provides a model together with the use of visualization
data to assess the efficiency of clustering by data experts.

157
 
Clustering

EXERCISE QUESTIONS

1. Briefly describe the K-Means Clustering algorithm.

2. Explain the difference between Manhattan Distance and Euclidean Distance calculations.
3. Explain how to interpret results from Centroid.
4. Give examples of problems suitable for data analysis with K-Means.
5. Search for a study using K-Means Clustering to analyze the data and summarize it briefly.

REFERENCES

Abbas, A., Prayitno, P., Butarbutar, F., Nurkim, N., Prumanto, D., Dewadi, F. M., Hidayati, N., & Windarto,
A. P. (2021). Implementation of clustering unsupervised learning using K-means mapping techniques.
IOP Conference Series. Materials Science and Engineering, 1088(1), 012004. doi:10.1088/1757-
899X/1088/1/012004
Bayram, E., & Nabiyev, V. (2020). Image segmentation by using K-means clustering algorithm in Eu-
clidean and Mahalanobis distance calculation in camouflage images. 2020 28th Signal Processing and
Communications Applications Conference (SIU), Signal Processing and Communications Applications
Conference (SIU), 2020 28th, 1–4.
Ben Ncir, C.-E., Hamza, A., & Bouaguel, W. (2021). Parallel and scalable Dunn Index for the validation
of big data clusters. Parallel Computing, 102, 102. doi:10.1016/j.parco.2021.102751
Erdoğmuş, Z. İ., & Arslan, M. K. (2022). Engaging with Social Media Influencers on Youtube: A Cluster
Analysis. Istanbul University Journal of the School of Business Administration, 51(1), 359–373.
Ginting, S. W., Hartati, R. S., Sudarma, M., & Swamardika, I. B. A. (2021). Clustering of Earthquake and
Volcanic Eruption Trauma Survivor Groups using K-Means Algorithm. 2021 International Conference
on Smart-Green Technology in Electrical and Information Systems (ICSGTEIS), Smart-Green Technol-
ogy in Electrical and Information Systems (ICSGTEIS), 2021 International Conference On, 69–73.
Govindaraj, V., Thiyagarajan, A., Rajasekaran, P., Zhang, Y., & Krishnasamy, R. (2020). Automated
unsupervised learning‐based clustering approach for effective anomaly detection in brain magnetic
resonance imaging (MRI). IET Image Processing, 14(14), 3516–3526. doi:10.1049/iet-ipr.2020.0597
Haviluddin, I. M., Putra, G. M., Puspitasari, N., Setyadi, H. J., Dwiyanto, F.A., Wibawa, A. P., & Alfred,
R. (2020). A Performance Comparison ofEuclidean, Manhattan and Minkowski Distances in K-Means
Clustering. 2020 6th International Conference on Science in Information Technology (ICSITech), Science
in Information Technology (ICSITech), 2020 6th International Conference On, 184–188.
Hyrmet, M., & Arbana, K. (2021). The Data Mining Approach: A Case Study - Clustering Algorithms
for After Sales Service. 10th Mediterranean Conference on Embedded Computing (MECO), 1-6.
Kumar, P., Ganapathy, G., & Kang, J.-J. (2021). A Hybrid Mod K-Means Clustering with Mod SVM
Algorithm to Enhance the Cancer Prediction. International Journal of Internet. Broadcasting and Com-
munication, 13(2), 231–243.

158

Clustering

Lou, N. (2022). Analysis of the Intelligent Tourism Route Planning Scheme Based on the Cluster Analy-
sis Algorithm. Computational Intelligence and Neuroscience, 2022, 1–10. doi:10.1155/2022/3310676
PMID:35800698
Munlika, H. (2019). Machine Learning with Clustering Model. Retrieved June 17, 2021, from https://me-
dium.com/tniuniversity/%E0%B8%81%E0%B8%B2%E0%B8%A3%E0%B8%97%E0%B8%B3-machine-
learning-%E0%B8%94%E0%B9%89%E0%B8%A7%E0%B8%A2-clustering-model-2a3c392e7faa
Peungvicha, P., Tubtiang, A. & Sattayaaphitan, T. (2019). Clustering of Jewellery Purchasing Behaviour
through Social Network. Journal of Humanities and Social Sciences, Rajapruk University, 5, 212-224.
Punhani, R., Arora, V. P. S., Sabitha, S., & Kumar Shukla, V. (2021). Applicationof Clustering Algorithm
for Effective Customer Segmentation in E-Commerce. 2021 International Conference on Computational
Intelligence and Knowledge Economy (ICCIKE), Computational Intelligence and Knowledge Economy
(ICCIKE), 2021 International Conference On, 149–154.
Puspasari, B. D., Damayanti, L. L., Pramono, A., & Darmawan, A. K. (2021). Implementation K-Means
Clustering Method in Job Recommendation System. 2021 7th International Conference on Electrical,
Electronics and Information Engineering (ICEEIE), Electrical, Electronics and Information Engineer-
ing (ICEEIE), 2021 7th International Conference On, 1–6.
Rezaeijo, S. M., Hashemi, B., Mofid, B., Bakhshandeh, M., Mahdavi, A., & Hashemi, M. S. (2021). The
feasibility of a dose painting procedure to treat prostate cancer based on mpMR images and hierarchi-
cal clustering. Radiation Oncology (London, England), 16(1), 1–16. doi:10.118613014-021-01906-2
PMID:34544468
Wafa, H. A., Aminuddin, R., Ibrahim, S., Mangshor, N. N. A., & Wahab, N. I. F. A. (2021). A Data
Visualization Framework during Pandemic using the Density-Based Spatial Clustering with Noise (DB-
SCAN) Machine Learning Model. 2021 IEEE 11th International Conference on System Engineering
and Technology (ICSET), System Engineering and Technology (ICSET), 2021 IEEE 11th International
Conference On, 1–6.
Wijaya, Y. A., Kurniady, D. A., Setyanto, E., Tarihoran, W. S., Rusmana, D., & Rahim, R. (2021). Da-
vies Bouldin Index Algorithm for Optimizing Clustering Case Studies Mapping School Facilities. TEM
Journal, 10(3), 1099–1103. doi:10.18421/TEM103-13
Xu, B., Chen, C., & Yang, J.-H. (2022). Application of Cluster Analysis Technology in Visualiza-
tion Research of Movie Review Data. Computational Intelligence and Neuroscience, 2022, 1–11.
doi:10.1155/2022/7756896 PMID:35880060
Zhou, Y., Yan, S., Wang, C., Zheng, K., & Zhu, L. (2021). Position Monitoring System Based on Hier-
archical Clustering. 2021 26th IEEE Asia-Pacific Conference on Communications (APCC), Communi-
cations (APCC), 2021 26th IEEE Asia-Pacific Conference On, 111–114.

159
160

Chapter 8
Association Rule

ABSTRACT
Businesses are increasingly seeking to understand consumer behavior and purchasing habits in order to
analyze the relationship between different products purchased by customers. By leveraging the associa-
tion rule mining technique, businesses can identify complementary products and bundle them together
to increase sales. This chapter provides an overview of association rule mining, a form of unsupervised
learning that enables data scientists to analyze the relationship between data items in a dataset. The
chapter explains the Apriori algorithm, a feature of association rule mining, and highlights how data
scientists can collaborate with domain experts to achieve business objectives such as product matching
and testing the efficiency of association rule results. Readers can follow a step-by-step guide to experience
the association rule mining technique using RapidMiner, enabling them to develop their understanding
of this valuable analytical tool.

INTRODUCTION

In business sectors, there is a need to sell multiple products at the same time. For example, superstores
often offer several products to their customers, such as offering soap along with shampoo. This kind of
product matching is done by analyzing the association rules on customer purchasing behavior (Ren, 2021:
Yingzhuo & Xuewen, 2021), which is the customers who buy soap often buy shampoo as well. There-
fore, association rules are used to process large data sets to analyze the correlation rules hidden within
the data (Paruechanon & Sriurai, 2019: Hong & Nan 2021: Salman & Sadkhan, 2021). The technique
is also used to analyze the correlation rules that can occur together; for example, when a car is broken,
it can lead to an accident while driving. In processing big data with the association rules, data scientists
may be given a number of rules. Data scientists, therefore, are necessary to analyze the benefits and ef-
ficiency of association rules in practical applications in industries. Data mining techniques are not only
used to analyze business data, but it has also been applied in many industries such as medical diagnosis,
Web Mining and Bioinformatics (Toumi, Gribaa, & Ben Abdessalem Karaa, 2021).

DOI: 10.4018/978-1-6684-4730-7.ch008

Copyright © 2023, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

Association Rule

ASSOCIATION RULE

In analyzing data with Association Rule, data scientists can use the following format (Zhang & Zhang,
2002).

Soap -> Shampoo

From such an association rule, it means that when customers buy soap, they will also buy shampoo.
Another metric used for association rule analysis is Confidence, which is the percentage of the availability
of rules-based data within the dataset. For example, if the Confidence value is 60 percent, and within
the dataset there are 30 instances of soap purchases out of the total 100 instances, it means there is 60
percent of shampoo purchase data, or there are 18 times that soap along with shampoo are purchased.
Another measure is Support, which refers to the percentage of compliant data in the entire dataset.
For example, a Support value of 2% means that the entire dataset has compliant instances, which is there
are 2 instances of buying soap along with shampoo out of all 100 instances.
To analyze association rules with Confidence and Support values, data scientists can calculate the
values of both as in the following example.
The store has a total of 100,000 customer purchases. Of all the data, the total number of milk purchases
is 40,000 times, and the total number of fruit purchases is 15,000. Of all the 40,000 milk purchases, the
number of purchases of milk along with fruit is 10,000 times. Confidence is calculated, and it results
as 25%, which means there is the frequency of buying fruit together with milk at 25 percent. And the
Support value is 10 percent which means that the frequency of buying fruit along with milk is at 10
percent of the total data. After analyzing the association rules, data scientists will obtain association
rules consisting of a large number of Confidence and Support values. Therefore, the data scientists need
to determine the Support and Confidence Threshold in order to measure whether each association rule
can be used in analysis. For example, when data scientists define Support and Confidence Thresholds
as 50%, after analyzing the data the Confidence and Confidence Thresholds must not be less than 50%
to be considered a Strong Association.
The data ready for analysis with association rules are usually in Boolean format. For example, to buy
equals 1, and not to buy equals 0. The analysis of association rules based on Boolean can be performed
as in the following example.

Table 1. Example of correlation rule analysis from Boolean data

Frequency of Each Purchase Soap Shampoo Detergent Dishwasher

1st Purchase 1 1 0 0
2 Purchase
nd
1 0 0 1
3 Purchase
rd
1 1 0 0
4th Purchase 1 1 1 1

161
 
Association Rule

From the table, the association rules can be analyzed with Confidence and Support values ​​as follows.

Soap -> Shampoo having 75% of Support value and 75% of Confidence value
Soap -> Dishwasher having 50% of Support value and 50% of Confidence value
Shampoo -> Detergent having 25% of Support value and 33% of Confidence value
Soap, Shampoo, Detergent -> Dishwasher having 25% of Support value and 100 percent of Confidence
value

From the association rules analyzed in the table, the number of Attributes is called Itemset. {Soap,
Shampoo} is a 2-Itemset. And {Soap, Shampoo, Detergent, Dishwasher}is a 4-Itemset.
Each dataset has a Frequency, or the itemset. For example, if Frequency is 2%, then the frequency of
{soap, shampoo}, which is 2-Itemset, is 2 times out of 100 total data. So, it is equal to Support, which
is also the frequency of association rules from all datasets. Data scientists need to determine the value
of Frequent Item Set, which should be greater than or equal to Support and Confidence Thresholds, so
that the data set contains sufficient Itemset to be used in the analysis with the association rule technique.

APRIORI ALGORITHM

With the Apriori algorithm, the data analysis process begins with the creation of all possible Frequent
Item Sets from the data set (Baby & Reddy, 2021: Hodijah & Setijohatmo, 2021). Then, only the Fre-
quent Item Sets with Confidence and Support values greater than or equal to Support and Confidence
Thresholds can be used to determine affinity rules. The procedure is as shown in the example.

Table 2. Data analysis sample set with Apriori algorithm

Frequency of Each Purchase Soap Shampoo Detergent Dishwasher Body Lotion

1st Purchase 1 0 0 1 1
2 Purchase
nd
0 0 1 1 0
3 Purchase
rd
1 0 0 1 0
4th Purchase 1 1 1 1 0
5th Purchase 1 1 0 1 0
6 Purchase
th
1 0 0 1 0
7 Purchase
th
0 0 1 1 0
8th Purchase 1 0 0 0 0
9th Purchase 1 1 0 1 0
10 Purchase
th
1 1 0 1 0

Based on the sample data set, the Support and Confidence Thresholds value is 20 percent when us-
ing Apriori Algorithm in association rule analysis. All possible correlation rules are analyzed from the
entire dataset starting with a 1-ItemSet with the frequency of each data as follows.

162

Association Rule

Table 3. Analysis of datasets starting with 1-ItemSet

ItemSet Frequency
Soap 0.8
Shampoo 0.4
Detergent 0.3
Dishwasher 0.9
Body Lotion 0.1

By setting the Support and Confidence Thresholds at 20 percent, the ItemSet consisting of Soap,
Shampoo, Detergent and Dishwashing Liquid has a Support and Confidence Thresholds value greater
than 20 percent, and can be created as a 1-ItemSet. In the next step, Apriori Algorithm will create 2-Item
data sets as follows.

Table 4. Analysis of the 2-ItemSet datasets

ItemSet Frequency
{Soap, Shampoo} 0.4
{Soap, Detergent} 0.1
{Soap, Dishwasher} 0.7
{Shampoo, Detergent} 0.1
{Shampoo, Dishwasher} 0.4
{Detergent, Dishwasher} 0.3

According to the Apriori Algorithm table, the frequency of the 2-ItemSet is generated from the
number of instances appearing in the frequency of two orders at the same time. For example, soap and
shampoo are brought together 4 times, so the frequency is 0.4. Only the 2-ItemSets can be created from
the 1-Itemsets with Support and Confidence Thresholds greater than 20%.
Then consider selecting only 2-ItemSet values ​​greater than 20% consisting of {soap, shampoo}
{soap, dishwasher} {shampoo, dishwasher} and {detergent, dishwasher}. Then Apriori Algorithm will
create 3-Item data sets.

Table 5. Analysis of the 3-ItemSet

ItemSet Frequency
{Soap, Shampoo, Dishwasher} 0.4

163
 
Association Rule

From the table, the 3-ItemSet is created by the frequency of appearing the 1-ItemSet within the
same instance up to 3 times, considering the first 1-ItemSet. Based on that dataset, the first 1-ItemSet is
Soap, and has 4 Instances consisting of Soap, Shampoo and Dishwasher, so Apriori Algorithm creates
a 3-ItemSet consisting of {soap, shampoo, Dishwasher} with a Support and Confidence Thresholds
greater than 20 percent.
The Apriori Algorithm then creates a 4-ItemSet. However, creating a 4-ItemSet starts with a 3-ItemSet
with the same first two data. But in that case, there is only one 3-ItemSet left, Apriori Algorithm will
stop creating ItemSet.
By creating the Frequent Item Set, Apriori Algorithm creates an association rule from a 3-ItemSet
by creating a Non-Empty Subset from a 3-ItemSet, and then creating an association rule. The Support
and Confidence Thresholds value is set as more than 55 percent.

Table 6. Creating association rules

Non-Empty Subset Association Rule Confidence Strong Association Rule

{Soap, Shampoo} Soap, Shampoo -> Dishwasher 0.4/0.4 = 1 Yes
{Soap, Dishwasher} Soap, Dishwasher -> Shampoo 0.4/0.7 = 0.57 Yes
{Shampoo, Dishwasher} Shampoo, Dishwasher -> Soap 0.4/0.4 = 1 Yes
{Soap} Soap -> Shampoo, Dishwasher 0.4/0.8 = 0.5 No
{Shampoo} Shampoo -> Soap, Dishwasher 0.4/0.4 = 1 Yes
{Dishwasher} Dishwasher -> Soap, Shampoo 0.4/0.9 = 0.44 No

From the table, the association rule is created from a 3-ItemSet. Firstly, all the data are taken to cre-
ate a total of 6 Subsets called Non-Empty Sets, and then the relationship rules are created from those
Subsets. For example, Sub set 1 consists of soap and shampoo. Then the association rule is created from
the data within the Set, which is dishwasher. The association rule is soap, shampoo -> dishwashing
liquid, which means every time the customers buy soap and shampoo, they will also buy dishwashing
liquid too. Confidence value of soap and shampoo is then analyzed from the frequency. If the frequency
of purchasing soap and shampoo at the same time is 4 times, and the frequency of purchasing soap,
shampoo and dishwashing liquid all together is 7 times, the Confidence value is 0.4 / 0.7 = 0.57, which
exceeds the Support and Confidence Thresholds value, being 55 percent. Therefore, the association rule
of soap, shampoo -> dishwashing liquid is a Strong Association Rule.
The association rules that could not be analyzed are soap -> shampoo, dishwasher, and dishwasher
-> soap, shampoo, which shows a Support and Confidence Thresholds below 55 percent.
By analyzing the data to create association rules with the Apriori Algorithm, the appropriate data for
analysis should be Categorial Data such as product type, gender and address. If data scientists use data
other than Categorial Data in Apriori Algorithm analysis, they must perform Attribute Transformation
to provide the data in the format suitable for data analysis with such techniques.

164

Association Rule

RULE INTERESTINGNESS

When data scientists perform association rule analysis from existing datasets, they need to consider the
benefits of analytical results in order to make decisions in business or industries (Bing, Yiming & Lee,
2001: Glass, 2013: Greco, Słowiński & Szczęch, 2016). The measurement can be divided into 2; 1.
Subjective Measure, which refers to the use of domain expert’s opinions or recommendations to decide
which association rules are useful for decision-making such as having physicians to decide on the asso-
ciation rules of cancer risk factor variables, 2. Objective Measure, which refers to the use of quantitative
analysis data to reflect the efficiency of the correlation rules obtained from data mining analysis. Data
scientists can use Objective Measures to analyze performance as follows:
The Support value is the frequency with which data appears in the entire dataset. Confidence value
is the frequency of having rules-compliant data within the dataset.
In cases where the Support and Confidence values ​​are greater than or equal to the Support and Con-
fidence Thresholds values, the association rules are considered appropriate to support decision-making.
In cases when data scientists obtain error during Data Pre-Processing. For example, Product A is sold
with Product B, resulting in every instance within the data set appearing the frequency of Product A
and Product B. The Confidence and Support values of A -> B are very high, but they do not have any
significant analysis because it is already a product that must be sold together.
Therefore, the metric that can solve this problem is Lift. Lift must be greater than 1, so that the rule
of relationship can be used to support decision making.

Table 7. Sample datasets for processing life values

Soap 1 1 1 1 0 0 0 0
Tooth Paste 1 1 0 0 0 0 0 0
Shampoo 0 1 1 1 1 1 1 1

From the table, you can create an association rule consisting of Support values ​​and Lift values as
follows.

Table 8. Association rules from the sample datasets

Association Rule Support Lift

Soap -> Tooth paste 25.00% 2.00
Soap -> Shampoo 37.50% 0.86
Tooth Paste -> Shampoo 12.50% 0.57

As in the table, the relationship rule of soap -> shampoo has a support value of 37.50%, and that of
toothpaste -> shampoo is at 12.50%.

165
 
Association Rule

However, the lift value of the relationship rule of soap -> shampoo is at 0.86, and that of the rela-
tionship of toothpaste -> shampoo is at 0.57. The lift values of both rules are less than 1, meaning it has
insignificant relationship to support decision. When looking back into the dataset, it has been found that
every time they buy soap or toothpaste, customers tend to buy shampoo as well. It means that regardless
the type of purchase, shampoo must be purchased too. Therefore, the aforementioned association rules
have no implications for supporting the decision because it is a fact that already exists.
Another metric that data scientists can use to consider in conjunction with lift is conviction. Convic-
tion should be greater than 1 as it is calculated as the Confidence of the first data divided by the Sup-
port of the second data. If the result is 1, it means the frequency that shows that the data from the entire
dataset of the first data is equal to the frequency of the second data in the association rules. Therefore,
it means that the two data are not related to each other. Data scientists can calculate Lift and Conviction
values as follows.
There is a sales dataset of 10,000 instances, which represents the frequency of 10,000 purchases.

Soap 4,000 pieces

Shampoo 8,000 pieces
The frequency of purchasing soap along with shampoo is 3,000 times.
The relationship rule has come out as Soap-> Shampoo
Support Value is 3,000 / 10,000 = 30%
Confidence Value is 3,000 / 4,000 = 75%

Lift value can be calculated by taking the Confidence value of the frequency of buying soap together
with shampoo divided by the Support value of the shampoo frequency, which is 0.75 / 0.8 = 0.94.
Conviction can be calculated by taking the Support value of the shampoo frequency that appears in
the entire data set, subtracting the number 1, and then dividing by the Confidence value of the frequency
soap purchase with shampoo and subtracting it by the number 1, which is (1 – 0.8) / (1 – 0.75) = 0.8.
Calculations show that if lift value is less than 1, this indicates that the association rule has no sig-
nificance for its use in decision support. Therefore, such association rules should not be used to make
business decisions.

ANALYZING ASSOCIATION RULES WITH RAPIDMINER SOFTWARE

In the data analysis experiment using the association rules, this book uses a sample dataset called shop-
ping.csv, which stores data on purchases. Data scientists have to import a data named shopping.csv into
RapidMiner software. The data objects within the dataset are also Boolean. True means Purchase, and
False means no purchase. The dataset includes 13 Attributes and 786 Instances.
Therefore, data scientists have to determine each attribute into Binomial values. In Unsupervised-
Learning processing, data scientists do not need to define the target variables as shown in the figure.
Meanwhile, the data scientists create the model by importing the shopping.csv dataset into the FP-
Growth model, which is used to analyze association rules aiming for creating a tree diagram, and reduce
the creation of a large number of Frequent Item Sets in Apriori algorithm. Then the min support value
is determined. RapidMiner is already set as 0.95, so reset it as 0.2 as shown in the picture.

166

Association Rule

Figure 1. Importing the shopping.csv dataset into RapidMiner software

In the figure, data scientists can connect a model from shoppoing.csv into the FP- Growth model,
and configure min support at 0.2.
In this figure, the association rule technique is analyzed using the shopping.csv dataset. RapidMiner
software has created Item Sets ranging from 1-ItemSet to 4-ItemSet. Every Item Set represents a Support
value, or the number of frequencies where the data appear.
However, when considering all the attributes within the data set, there is still the data irrelevant with
the frequency of customer purchases. Data scientists have to connect to Select Attributes model in order
to select specific attributes related to the frequency of customer purchases.
Data scientists use the Select Attribute model to select only Attributes related to the frequency of
customer purchases by specifying the Parameter as a Subset and then selecting Select Attributes by
selecting the Attributes; Children, Gender and Working into the box on the right.
After defining the parameters, the data scientists can select Invert Selection so that the RapidMiner
software knows it is to use the remaining Attributes that are not the ones chosen for the data analysis.
The result after selecting Attribute is as follows.
After the data scientists have selected only the attributes needed to analyze the data, RapidMiner can
use the association rule technique to create more Frequent Item Sets, being up to 5-ItemSet.
Once the Frequent Item Set has been created, the data scientists will take all Item Sets to create as-
sociation rules using the Create Association Rules model, which connects the FP-Growth model to the
association rule display as shown below.

167
 
Association Rule

Figure 2. Beginning data analysis with the association rule technique

To use the Create Association Rules model to create association rules, data scientists can choose a
measure of the efficiency of the strong and valuable association rules to support decision making by
defining Parameters as follows.
From the figure, the data scientists can select measures such as Confidence, Life, and Conviction.
The minimum value of min confidence can be set. The result is as follows.
From the figure, the data scientists can determine the outcome of the rule. For example, in Associa-
tion Rule No. 5, when a customer buys Fresh Vegetables, they also buy Tinned Goods. The Confidence
value is 0.81, which is considered a reliable relationship rule based on a quantitative measure.

168

Association Rule

Figure 3. Results of data analysis by association rule technique

Figure 4. Selecting only the Attribute needed for data analysis

169
 
Association Rule

Figure 5. Determining parameters to select attribute models

Figure 6. Results after selecting attributes

170

Association Rule

Figure 7. Creating association rules using create association rules

Figure 8. Selecting a metric for measuring the efficiency of the association rules

CASE STUDY

Using Data Result Analyzed With Association Rule Mining to Expand the
Lifetime Value of Customers in Banking Services (Rezaei et al., 2022)

We are living in an era where the popularity of using each customer’s products and services has been
much shorter. There are many emerging goods and services that have become popular in their use and are
rapidly deteriorating in popularity. Customer Life Time Value: CTV becomes the data that each business

171
 
Association Rule

Figure 9. Results from creating association rules

needs to analyze to identify which groups of customers will return to buy to create value for the business
so that businesses can create marketing strategies specific to these valuable customers. Technological
changes have also resulted in the customer service center becoming a repair and maintenance center
due to the fact that customers can visit products through the website and place orders online. The use of
internet technology has destroyed international borders, and businesses are expanding to their customers
worldwide. On the other hand, customers also need stores that have the ability to display their products
and support their purchases via mobile devices such as Smart Phones or Tablets.
The change directly impacts the banking business, which have to provide a variety of digital channels
to support the changing style of the service recipients, such as the provision of web-based applications
that support the client’s financial transactions, the provision of native apps installed on mobile phones
to support the client’s financial transactions via mobile phones. However, those marketing strategies
are all so similar that the bank is unable to communicate the right marketing for the customer segments
on each channel. The marketers are unable to analyze the behavior of customers on each channel and
develop marketing strategies that can increase customer life time value.
The Omnichannel has become a way to get through this problem by combining business operations
and online-offline marketing strategies. Such business channels have enabled marketers to identify market
segmentation that overlaps between customers with consumption habits through online and offline chan-
nels. Business operations through such channels also have a similar purpose to other forms of business
operations; to try to get consumers to turn their attention to their own products and services, maintain-
ing an existing customer base, analyzing marketing data to uncover customer behavior and products or
services that customers are interested in, and prolonging the time the customers stay interested in their
own business by trying to promote marketing or create after-sales service that keeps customers coming
back for more services.
From the transition to providing services in an online format, banking businesses have reduced the
number of branches and staff in line with the popularity of online transactions by customers in order to
prevent customers from moving to other banking services. The banks need to conduct a study on customer
lifetime value based on three factors; When did the customer last use the banking service? How many
times did the customer use the banking service? How much money have the customers spent through

172

Association Rule

the bank? These three factors are variables to determine the customer lifetime value so that the bank
can determine whether the customer will return to the service or change their mind to other banking
services. This is the era where mobile banking is widely used among customers to transfer money, pay
for goods and services to retail businesses or merchants.
In 2022, a research was conducted on the analysis of the relationship between Customer Life Time
Value and omnichannel business practices (Rezaei, et. al, 2022). The research put forward a case study
of agricultural banking, in which services are provided to customers both offline via bank branches and
online format via mobile banking. The bank itself has made efforts to assess the satisfaction of service
recipients in every touchpoint of the service process to improve the level of satisfaction and delivers a
good experience to their customers. This effort has led to the issue of analyzing whether the Customer
Life Time Value of omnichannel business is different from other business models. In this case study
of the Agricultural Bank, the research framework environment is different from other types of banking
business because the Agricultural Bank is established as a state-owned enterprise which has a higher
authority to govern.
In the data exploration phase, the researchers used the data source from the transactions of customers
who have used banking services for the past 2 years, then used the Association Rule Mining Technique
to analyze the relationships. The relationship rules can then be analyzed as 9 types. The overall pattern
revealed that the customers using the “Point of Sale” system (a retail payment system through a digital
device) tend to be used to the ATM system. And lastly are the transactions at a bank branch in case the
“Point of Sale” system and ATMs cannot support such transactions. In the analysis of such relationship
rules, another form of relationship was found. It is the group of customers who did not receive service
from the bank in the form of omnichannel. The uses only the “Point of Sale” system, not multi-channels.
Therefore, service recipients can be classified into two types; Omnichannel and Non-Omnichannel. The
researcher then used the T-Test Technique to prove the difference of Customer Life Time Value between
the two service recipients and found that the Customer Life Time Value between these two groups is
significantly different.
By segmenting customers by relationship rules model and testing the difference between Customer
Life Time of customers with omnichannel and non-omnichannel relationship rules behaviors, Bank for
Agriculture and Agricultural Cooperatives are cost-effective to provide digital channels along with the
provision of financial services at branches because there are still groups of customers who are loyal to
the bank from receiving services in multiple channels at the same time. This research suggests using om-
nichannel operations to extend the Customer Life Time Value by retaining existing customers to continue
to buy products and services of the business without switching to competitors’ products and services.

Analysis of Language Courses Based on Learner Behavior With

Association Rule Mining (Chaimongkol & Meesad, 2010)

Many language schools try to provide language courses that stand out from other competing language
schools. As a result, those language schools have sought to study the needs of learners to develop courses
that can be used to attract those who want to improve themselves in the field of foreign languages. They
can also build the loyalty of current learners to resume classes or take other courses provided by the
school. The challenge in preparing those courses is that the needs of individual learners are diverse and
difficult to analyze. And it is even more difficult to associate those patterns with the language courses
that are appropriate for each learner. Therefore, the data mining technique is one of the ways to use the

173
 
Association Rule

information that each school has to analyze the patterns of co-learners and analyze the linkages between
the behaviors and the courses that the school currently has. To analyze such data, language schools can
use data scientists to apply various forms of data mining techniques such as decision trees, neural net-
works, or even association rule mining techniques to analyze the data.
In 2010, there was a research attempting to analyze the correlation between learners’ learning behaviors
in relation to their choice of language courses (Chaimongkol & Meesad, 2010). The research extracted
data about the actual needs of learners by dividing them into the fundamental needs for language use
and the need to enhance their ability to use a foreign language. Data scientists have designed processes
for data processing including: 1. Surveying datasets that can be used for data analysis with data mining
techniques to answer the question the language schools intend to know, 2. Preparing data in a format
that is ready for processing with data mining techniques; normalization or data engineering, 3. Creating
models for analyzing the rules, the relationship between learner needs patterns, and language courses,
and 4. Using the results from the analysis of data in the form of correlation rules to translate the results.
From such process, data scientists surveyed and collected data on language courses for students in
Thailand from four language schools from January 2007 to January 2008. The features are the courses in
which there are Data Objects such as Junior Courses or Advance Courses, the time of day that students
choose to study, in which there are Data Objects such as 8.00 a.m., the number of registration hours in
which there are Data Objects such as 400 hours and the number of days students choose to study each
week in which there are Data Objects such as 3 days per week. The result of the data analysis is a set of
correlation rules. For example, language schools should set the frequency of Junior Course at 3 hours
per week.
From the case studies, businesses can use Association Rule Mining to analyze data to uncover hid-
den relationships in datasets, which leads to designing products and services that can meet different
customer expectations.

CONCLUSION

Data scientists can use RapidMiner software to analyze datasets with association rules. Data suitable
for analysis by such technique should be Category Data such as product type, gender and address. The
association rules aim to analyze the correlation of data from the frequency of occurrence of such data to
show the results of their relationship. For example, those who buy wines tend to also buy cheese. Users
can apply the rules of association in business and other industry operations. However, in the process of
establishing the association rules, there will appear many rules that the data scientists must consider as a
strong rule that can be used for decision making by considering quantitative metrics such as Confidence,
Lift and Conviction, as well as consulting Knowledge Experts. This will present the association rules
that are factual and beneficial to the organizations.

EXERCISE QUESTIONS

1. Describe the purposes of data analysis with association rules.

2. Explain the difference between Confident and Support.
3. Explain the meaning of Frequency Itemsets.

174

Association Rule

4. Give an example of the problem that requires the association rules in data analysis.
5. Find a research paper that uses association rules in data analysis and summarize it briefly.

REFERENCES

Baby, S. S., & Reddy, S. L. (2021). End to End Product Recommendation system with improvements
in Apriori Algorithm. 2021 Third International Conference on Inventive Research in Computing Appli-
cations (ICIRCA), Inventive Research in Computing Applications (ICIRCA), 2021 Third International
Conference On, 1357–1361.
Bing, L., Yiming, M., & Lee, R. (2001). Analyzing the interestingness of association rules from the
temporal dimension. Proceedings 2001 IEEE International Conference on Data Mining, Data Mining,
2001. ICDM 2001, Proceedings IEEE International Conference on, Data Mining, 377–384. 10.1109/
ICDM.2001.989542
Chaimongkol, N., & Meesad, P. (2010). Association Rule Mining for Specific New Course. Mahanakorn
University of Technology (MUT).
Glass, D. H. (2013). Confirmation measures of association rule interestingness. Knowledge-Based Sys-
tems, 44, 65–77. doi:10.1016/j.knosys.2013.01.021
Greco, S., Słowiński, R., & Szczęch, I. (2016). Measures of rule interestingness in various perspectives
of confirmation. Information Sciences, 346–347, 216–235. doi:10.1016/j.ins.2016.01.056
Hodijah, A., & Setijohatmo, U. T. (2021). Analysis of frequent itemset generation based on trie data
structure in Apriori algorithm. Telkomnika, 19(5), 1553–1564. doi:10.12928/telkomnika.v19i5.19273
Hong, G., & Nan, G. (2021). Research and Application of a Multidimensional Association Rules Mining
Algorithm Based on Hadoop. 2021 IEEE Intl Conf on Parallel & Distributed Processing with Applica-
tions, Big Data & Cloud Computing, Sustainable Computing & Communications, Social Computing &
Networking (ISPA/BDCloud/SocialCom/SustainCom), Parallel & Distributed Processing with Applica-
tions, Big Data & Cloud Computing, Sustainable Computing & Communications, Social Computing
& Networking (ISPA/BDCloud/SocialCom/SustainCom), 2021 IEEE Intl Conf on, ISPA-BDCLOUD-
SOCIALCOM-SUSTAINCOM, 636–643.
Paruechanon, P., & Sriurai, W. (2019). Applying association rule to risk analysis for student-dropout in
Information Technology Program. Journal of Science and Science Education, 1(2), 123-133.
Ren, X. (2021). Application of Apriori Association Rules Algorithm to Data Mining Technology to Mining
E-commerce Potential Customers. 2021 International Wireless Communications and Mobile Computing
(IWCMC), Wireless Communications and Mobile Computing (IWCMC), 2021 International, 1193–1196.
Rezaei, M., Sanayei, A., Aghdaie, S. F. A., & Ansari, A. (2022). Improving the Omnichannel Custom-
ers’ Lifetime Value Using Association Rules Data Mining: A Case Study of Agriculture Bank of Iran.
Iranian Journal of Management Studies, 15(1), 49–68.

175
 
Association Rule

Salman, W. A. K., & Sadkhan, S. B. (2021). Proposed Association Rules Mining Algorithm for Sen-
sors Data Streams. 2021 1st Babylon International Conference on Information Technology and Science
(BICITS), Information Technology and Science (BICITS), 2021 1st Babylon International Conference
On, 76–81.
Toumi, A., Gribaa, N., & Ben Abdessalem Karaa, W. (2021). Mining biomedical texts based on statistical
method and association rules. 2021 International Conference of Women in Data Science at Taif University
(WiDSTaif), Women in Data Science at Taif University (WiDSTaif), 2021 International Conference Of, 1–4.
Yingzhuo, X., & Xuewen, W. (2021). Research on Community Consumer Behavior Based on Associa-
tion Rules Analysis. 2021 6th International Conference on Intelligent Computing and Signal Processing
(ICSP), Intelligent Computing and Signal Processing (ICSP), 2021 6th International Conference On,
1213–1216.
Zhang, C., & Zhang, S. (2002). Association Rule Mining. In Models and Algorithms. Springer Berlin
Heidelberg. doi:10.1007/3-540-46027-6

176
 177

Chapter 9
Recommendation System

ABSTRACT
Recommendation systems are critical tools used by marketing departments to provide customers with
product recommendations. Data scientists also use recommendation system analysis to assess the effec-
tiveness of product and service suggestions. There are two types of recommendation systems: content-
based and collaborative filtering. Content-based recommendations are based on the customer’s purchase
history, while collaborative filtering suggests a product based on purchasing behavior. Collaborative
filtering can be divided into content-based filtering, which suggests products based on similar purchas-
ing behavior, and item-based filtering, which suggests products based on their attributes. User-based
and item-based nearest-neighbor collaborative filtering and probabilistic methods are used to analyze
data and provide product suggestions. Businesses rely on recommendation systems to achieve various
objectives such as customer retention and increased ROI. RapidMiner can analyze data following the
principles of the recommendation system, as demonstrated in this chapter, step-by-step.

INTRODUCTION

Many organizations have a set of data that can be used to support decision-making on matters such as
marketing data. Such data are therefore used for the creation of an advisory system. For example, when
a customer logs into an online bookstore, the recommendation system will offer books that are similar
to the ones the customers have previously ordered. This is to increase the chances of a customer pur-
chase. Recommendation System can increase a lot of sales for online stores. It is widely used in many
industries such as tourism and hotels where the system advises on accommodation and attractions in line
with the tastes of the customers (Aditya et al, 2021: Chen & Qin, 2021: Wayan Priscila Yuni Praditya,
Erna Permanasari, Hidayah, 2021).

DOI: 10.4018/978-1-6684-4730-7.ch009

Copyright © 2023, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
 
Recommendation System

RECOMMENDATION SYSTEM

The recommendation system is divided into two types.

1. Content-Based Recommendation: It focuses on making recommendations using past content as a

reference (Tai, Sun & Yao 2021). For example, when a customer has previously purchased Product
A, the system will present Product A and other products that are similar in properties to Product
A.
2. Collaborative Filtering: CF focuses on analyzing customer behavior (Emon et al, 2021: Khatter
et al, 2021). For example, Customer A tends to buy soap and shampoo all together, so Customer
B who buys soap will likely buy shampoo too. The aforementioned system is also divided into 2
forms.
a. Content-Based Filtering: This is to focus on analyzing user behavior such as a group of cus-
tomers with similar purchasing behavior (Yadalam et al., 2020: Kamukwamba & Chunxiao,
2021).
b. Item-Based Filtering: This is to focus on analyzing product properties (Kusumawardhani,
Nasrun & Setianingsih, 2019: Lourenco & Varde 2020: Padhy, Singh & Vetrivelan, 2022).
For example, scooter and skateboards are different products but share similar properties.

The CF processing algorithm is a prediction of the similar data from the existing data, as in the example.

Table 1. Examples of datasets for forecasting data with recommendation system

Data Mining Technology

Python Design Thinking
Techniques Management
Sarawut Ramjan 5 2 5 4
Jirapon Sunkpho 2 5 ? 3
Kom Campiranon 2 2 4 2
Purimprach Saengkaew 5 1 5 ?

Sample dataset presents the student’s satisfaction on each subject. 1 means most disliked while 5
means most liked. In CF processing algorithms, data scientists put datasets into processing to predict
empty values. For example, Jirapon Sunkpho has not filled out his satisfaction in the Design Thinking
course, so CF processing will use existing data to predict what extent Jirapon Sunkpho is likely to favor
the course. If the forecast result is at level 5, it means most favorable. The recommendation system will
recommend Design Thinking course to Jirapon Sunkpho.
Data mining techniques applied to the development of the recommendation system can be divided
into 2 types: 1. Non-Probabilistic Algorithms such as K-Nearest Neighbors (KNN); both User-Based
and Item-Base, and 2. Probabilistic Algorithms such as Bayesian-Network Modes.

178

Recommendation System

USER-BASED NEAREST-NEIGHBOR COLLABORATIVE FILTERING

The fundamental principle of User-Based Nearest-Neighbor Collaborative Filtering is to forecast the

current missing data by considering other data with a complete similar format. It starts with analyzing the
similarity of the datasets, which are human opinions, and then uses that rich data to forecast the current
missing data (Adeniyi, Wei & Yongquan, 2016: Riswanto, Suparyanto & Robi’In, 2019). For example,
if Mr. A has evaluated his liking on the movies “Bangkok Traffic (love) Story” and “Hormones”, but has
not evaluated his liking of the Thai movie Nang Nak, the User-Based Nearest-Neighbor Collaborative
Filtering will bring the data of others such as Mr. B, who has evaluated his liking for the movie “Bangkok
Traffic (love) Story”, “Hormones” and “Nang Nak” to predict Mr. A satisfaction on “Nang Nak” movie.
In this example, User-Based Nearest-Neighbor Collaborative Filtering uses the similarity of Mr. A
and Mr. B’s data models to predict the missing data. The significant principle of this is to calculate the
similarity of the two datasets to forecast the missing data.

MEASURING USER SIMILARITY: PERSON CORRELATION

Pearson Correlation is an analysis of the relationship and similarity between two datasets with values​​
between – 1 and 1. If the value obtained from the analysis is minus, it means the relationship between the
two datasets is inverse. For example, the more Mr. A likes the movie Nang Nak, the more Mr. B tends
to like the movie too. In other words, if the data obtained from the analysis is positive, it means that the
higher the value of one dataset, the more valuable the other dataset. By Pearson Correlation method, the
similarity is calculated as follows.

a,b : Data Scientists’ interests in complementing Data

ra ,p : Data that require the complement or p of the dataset a
ra : average of dataset a
rb : average of dataset b
P : datasets a and b

∑ p P (ra ,p − ra ) (rb,p − rb )

sim (a, b ) = 
∑ p P (ra ,p − ra ) ∑ p P (ra ,p − ra )
2 2

a is the data rating the liking for each movie, but does not assess the liking of the movie Nang Nak.
So, data scientists have to calculate similarity from other datasets to predict the liking for Nang Nak of
dataset a, as shown below.
From the table, data scientists can do the calculations to find similarity between Sarawut Ramjan’s
fondness on each movie and of others with the use of Pearson Correlation as follows.

(5 −r ) *(3 −r ) +( 3 −r ) *(1 −r ) + … + (4 −r ) *(3 −r )
a b a b a b
= 0.85
(5 −r ) + (3 −rb ) + … + (4 −ra ) +(3 −rb )
2 2 2 2
a

179
 
Recommendation System

Table 2. Examples of datasets for forecasting data using Pearson correlation

The Chains
Hormones The Slavery To Irate, to Love Nang Nak
of Sin
Sarawut Ramjan 5 3 4 4 ?
Jirapon Sunkpho 3 1 2 3 3
Kom Campiranon 4 3 4 3 5
Purimprach Saengkaew 3 3 1 5 4
Tanatorn Tanantong 1 5 5 2 1

From the calculation with Pearson Correlation, which evaluates Sarawut Ramjan and Jirapon Sunk-
pho’s fondness on the movies Hormones, The Slavery, To Irate-To love and The Chains of Sin without
taking the results from the movie Nang Nak to calculate. The similarity is at 0.85. Data scientists can
take the level of Sarawut Ramjan’s fondness to calculate with Kom Campiranon, Purimprach Sangkaew
and Tanatorn Tanantong’s fondness, and the results are 0, 0.7 and – 0.79, respectively.
Therefore, the level of Jirapon Sunkpho ‘s fondness on the movies is most similar to Sarawut Ram-
jan’s. Afterward, data scientists can predict the fondness level Sarawut Ramjan has on Nang Nak movie,
which can be selected from Jirapon Sunkpho and Purimprach Sangkaew’s fondness data on the movies.
The similarity values are at 0.85 and 0.7, which are close to 1. It is to say that the more fondness Jirapon
Sunkpho and Purimprach Sangkaew have for the movies, the more Sarawut Ramjan tends to like the
movies. The forecast can be made using the following formula.

∑ p NSim (a, b ) (rb,p −rb )

pred (a, p ) = ra
∑ p NSim (a, b )

From the formula, scientists can predict the satisfaction Sarawut Ramjan has on Nang Nak from the
liking of Jirapon Sunkpho and Purimpan Sangkaew as follows.

0.85 * (3 − 2.4) + 0.7 * (4 − 3.8)

4+ = 4.42
0.85 + 0.7

Sarawut Ramjan’s fondness on Nak Nak is 4.42, which means that the data scientists can fill in the
missing values of the data set using other similar data. In this case, the scientists could choose to use the
data from Jirapon Sunkpho, which shows the greatest degree of similarity without using any other dataset.

ITEM-BASED NEAREST NEIGHBOR RECOMMENDATION

When analyzing data with User-Based Nearest-Neighbor Collaborative Filtering, the greater the amount
of data, the more digital resources the scientists need to prepare; memory and processing time. Therefore,
Item-Based Nearest Neighbor Recommendation will not analyze the similarity which is the opinion of

180

Recommendation System

humans, but it analyzes the similarity of the properties such as product similarity (Gong, Ye, H-W, &
Zhu, 2009). The calculation formula is as follows.

 
a , b : Item on which the scientists need to complement the data
ru,a : Data in Item a
ru,b : Data in Item b
ru : Mean of Each Item
U : Number of customers

  ∑ pU (ru,a −ru ) (ru,b −ru )

( )
sim a , b =
∑ pU (ru,a −ru ) ∑ pU (ru,b −ru )
2 2

The Item-Based Nearest Neighbor Recommendation analysis aims to take the weights the customers
give to each item to subtract the average of each customer. The data can be analyzed as in the following
example.

Table 3. Datasets for data forecasting using item-based nearest neighbor recommendation

Bathroom
Soap Toothpaste Shampoo Dishwasher
Cleaner
Sarawut Ramjan 5 3 4 4 ?
Jirapon Sunkpho 3 1 2 3 3
Kom Campiranon 4 3 4 3 5
Purimprach Saengkaew 3 3 1 5 4
Tanatorn Tanantong 1 5 5 2 1

Data scientists can analyze the similarity of bathroom cleaner with other products to fill in the miss-
ing data by using a similar model to forecast. The data that customers give each item weight are taken to
subtract with the average value of each customer. For example, Sarawut Ramjan gives the soap weight,
from scale1 to 5 of satisfaction, and he also gives other products weight; the average weight is 4. So, the
difference (Mean-Adjusted Ratings Database) that Sarawut Ramjan has on soap is 1. When calculating
the differences in every customer, the results are as follows:
When receiving the difference, the scientists can use data from Item 1, which is soap, to calculate
the similarity with Item 5, the toilet cleaner as follows.

0.6 * 0.6 + 0.2 * 1.2 + (−0.2) * 0.80 + (−1.8) * (−1.8)

= 0.80
0.62 + 0.22 + (−0.2) + (−1.8) * 0.62 + 0.22 + 0.82 + (−1.8)
2 2 2

181
 
Recommendation System

Table 4. Mean-adjusted ratings database

Bathroom
Soap Toothpaste Shampoo Dish Washer
Cleaner
Sarawut Ramjan 1.00 -1.00 0.00 0.00 ?
Jirapon Sunkpho 0.60 -1.40 -0.40 0.60 0.60
Kom Campiranon 0.20 -0.80 0.20 -0.80 1.20
Purimprach Saengkaew -0.20 -0.20 -2.20 2.80 0.80
Tanatorn Tanantong -1.80 2.20 2.20 -0.80 -1.80

In the case soap has similar data format to toilet cleaner, the scientists can make predictions to
complement data.
∑ i ratedItem (u ) Sim (i, p ) : The sum of the similarities between the item used in the calculation
and the item to forecast data

ru,i : The weight the customer gives to the item, used in the calculation

∑ i ratedItem (u ) Sim (i, p ) * ru,i

pred (u, p ) =
∑ i ratedItem (u ) Sim (i, p )

According to the formula, the scientists can predict the missing data of Sarawut Ramjan’s weight on
bathroom cleaner as in the following calculations.

sim (1, 5) * 5 + sim (2, 5) * 3 + sim (3, 5) * 4 + sim (4, 5) * 4

prad (Sarawut Ramjan’s,Item 5) =
sim (1, 5) + sim (2, 5) + sim (3, 5) + sim (4, 5)

Based on the calculations, the similarity between Item 1 – 4 and Item 5, along with the weight that
Sarawut Ramjan gives to every Item, is used in the calculations. The similarity of Item 1 and 5 is then
multiplied by the number 5, which Sarawut Ramjan has assessed, and then divided by similarity values
of Item 1 – 4 and Item 5.

PROBABILISTIC METHODS

Probabilistic Methods is one of the methods to predict the missing data. For example, the ratings that
moviegoers would likely have for a particular movie can be predicted using the Bayes Theorem technique.
The use of such techniques is based on calculating probability of a preceding event as in the following
formula.

Y = The probability that event Y will occur

X = events that have already occurred

182

Recommendation System

P (Y|X ) * P (Y )
P (Y|X ) =
P (X )

The data on the probability of an event to occur are as follows.

Table 5. Examples of data sets for calculating probabilities

The Chains
Hormones The Slavery To Irate-To Love Nang Nak
of Sin
Sarawut Ramjan 1 3 3 2 ?
Jirapon Sunkpho 2 4 2 2 4
Kom Campiranon 1 3 3 5 1
Purimprach Saengkaew 4 5 2 3 3
Tanatorn Tanantong 1 1 5 2 1

Data scientists can calculate all cases, and the movie viewers can evaluate their liking on each movie
from level 1 to 5. Therefore, the chance that Sarawut Ramjan will evaluate the film Nang Nak can be
in 5 cases.

P(X|Item5 = 1)
= P(Item1 = 1|Item 5 = 1) * P(Item2 = 3| Item 5 = 1)
* P(Item3 = 3 | Item 5 = 1) * P(Item 4 = 2 | Item 5 = 1)

2 1 1 1 1
= * * * * » 0.125
2 2 2 2 2

The result of the calculation in the case of Item 5 is 1, which means the probability that Sarawut
Ramjan will evaluate his fondness on the movie Nang Nak is at level 1, calculated from the proportion
of the fondness level in Item 5 and other items; for example, P (Item1 = 1| Item 5 = 1). When consider-
ing the data of Item 5, there are 2 users giving level1 liking, so the denominator is 2. And there are 2
Item1, so the numerator is 2, which is 2/2. By considering the proportion of Item5 and other Items, and
multiplying all together, so the probability is calculated as 0.125.
Data scientists need to calculate such probability in all cases from 1 to 5, and the probabilities are
then applied to predict the missing data.

EVALUATION MATRIX

Data scientists can evaluate the efficiency of data forecasting through the recommendation system which
consists of

183
 
Recommendation System

1. Predict Accuracy: It is the difference between the data used for teaching machines and the data
received from the forecast. The lower the Root Absolute Error value, the more accurate the value
obtained from the forecast. In the Recommendation System, the Evaluation Matrix from Predict
Accuracy can be used to measure the efficiency and accuracy of the recommendation.
2. Rank Accuracy: This is the percentage of precision of the Recommendation System. For example,
the system recommends 5 movies, and there are 4 movies that the user likes. This means that the
percentage of recommendation accuracy is 80 percent.

RECOMMENDATION SYSTEM WITH RAPIDMINER

Data scientists can analyze the Recommendation System with RapidMiner software by downloading ad-
ditional installations at Extensions and selecting Marketplace (Updates and Extensions) as shown below.

Figure 1. Accessing marketplace to download additional installation

Next, the scientists can search for the Recommender Extension, select for Installation, and press the
Install Packages button.
During the software installation process, RapidMiner will ask for license confirmation. The data
scientists can select I accept the terms of all license agreements and then select Install packages.
After the installation, data scientists can Import Ratings.csv dataset into RapidMiner software.
Within the dataset, it contains data on the fondness level each reader has on the books, totaling 10,000
instances as follows.
Afterward, the scientists can set role of each attribute, so that they can import and process with Item
Recommendation as follows.
The data scientists can select Edit List (2) to Set Roles for the data within the dataset.
In the figure, the data scientists need to define rating as a label or target variable, set user_id as user
identification, and set book_id as item identification, in order that such data can be processed with the
Item Recommendation model.
For data recommendation, data scientists have to divide the data into two parts according to the
principle of Supervised-Learning using the Split Data model.
In the figure, data scientists use Split Data to divide the data into two parts; 80% for teaching machine
learning, and 20% for testing the model performance.
From the picture, the scientists can press Edit Enumeration button (2) and then determine the pro-
portion to 0.8, which is for teaching the machine learning, and 0.2 as the proportion to test the model.

184

Recommendation System

Figure 2. Recommender extension installation

In the figure, data scientists connect Split Data model, 80% of which is designated for teaching machine
learning, to the User K-NN under the Recommender implementation in order to forecast the missing
data. Then 20% of the data used for the test is connected to the data obtained from the forecast, and put
them into the Apply Model (Item Recommendation) for analysis using Recommendation System. The
results of data analysis are as presented in the picture.
Recommendation System can recommend books for each reader. For example, the system recom-
mends 100 books to Reader 2 based on user_id number 2 and the number of books in the Item_id that
the model has suggested. Then the efficiency of the recommendation can be tested using the Performance
(Item Recommendation) model as follows.
Data scientists can test the efficiency of the system by connecting the machine-learning model from
K-NN learning into the Performance model, then connecting the forecasted results from User K-NN to
the Performance train. Then, 20 percent for testing the data is taken into Test of Performance to analyze.
The efficiency testing results are as in the table below.
The result of test using a model within the RapidMiner software shows Rank Accuracy, which is
the percentage of accuracy that the Recommendation System provides to the users. For example, if the
prec@5 value is 0.137, it means the accuracy is 20 percent.

185
 
Recommendation System

Figure 3. License agreement

CONCLUSION

Data scientists can develop models to provide called Recommendation System, both in the form of User-
Based Recommendation which analyzes similar behaviors of each user in order to recommend products
or services consistent with customer behavior, and the form of Item-Based Recommendation which
focuses on the similarity of the products. Both aspects are aimed at forecasting of the missing data. The
principle of data forecasting is consistent with the concept of Supervised-Learning. Data scientists still
have to divide the data into two parts, for teaching machines and for testing the performance of models.
Using RapidMiner software for recommending data under the concept of Recommendation System,
data scientists can use models in the Recommender Package that provides data aligned with the con-
cept of Recommendation System such as User K-NN, Apply Model and Performance. All of them are
Item-Based Recommendation data analysis. Data scientists can analyze computing efficiency in a Rank
Accuracy manner, the proportion of accurate predictions or recommendations.

186

Recommendation System

Figure 4. Importing ratings dataset into RapidMiner software

Figure 5. Setting roles for the dataset

187
 
Recommendation System

Figure 6. Setting roles on data for item recommendation processing

Figure 7. The data is divided into two sections: for teaching machine learning and for testing the model

EXERCISE QUESTIONS

1. Explain the concept of Content Based Recommendation.

2. Explain the concept of Collaborative Filtering
3. Explain the difference between Content-Based Filtering and Item-Based Filter.
4. Give examples of problems that require Recommendation System to provide users with recom-
mendations and answers.
5. Search for the research that uses Recommendation System, and summarize it.

188

Recommendation System

Figure 8. Dividing the dataset into two parts using split data

Figure 9. Forecasting and testing data for recommendation

189
 
Recommendation System

Figure 10. Results from data analysis using recommendation system

Figure 11. Testing the efficiency of the recommendation system using performance

Figure 12. Results of the recommendation system efficiency

190

Recommendation System

REFERENCES

Adeniyi, D. A., Wei, Z., & Yongquan, Y. (2016). Automated web usage data mining and recommenda-
tion system using K-Nearest Neighbor (KNN) classification method. Applied Computing & Informatics,
12(1), 90–108. doi:10.1016/j.aci.2014.10.001
Aditya, G. M., Hoode, A., Anvesh Rai, K., Biradar, G., Ajay Kumara, M., Manoj Kumar, M. V., Prashanth,
B. S., Sneha, H. R., & Shivadarshan, S. L. (2021). Machine Learning Based Platform and Recommenda-
tion System for Food Ordering Services within Premises. 2021 2nd Global Conference for Advancement
in Technology (GCAT), Advancement in Technology (GCAT), 2021 2nd Global Conference For, 1–8.
Chen, Q., & Qin, J. (2021). Research and implementation of movie recommendation system based
on deep learning. 2021 IEEE International Conference on Computer Science, Electronic Information
Engineering and Intelligent Control Technology (CEI), Computer Science, Electronic Information En-
gineering and Intelligent Control Technology (CEI), 2021 IEEE International Conference On, 225–228.
Emon, M. I., Shahiduzzaman, M., Rakib, M. R. H., Shathee, M. S. A., Saha, S., Kamran, M. N., &
Fahim, J. H. (2021). Profile Based Course Recommendation System Using Association Rule Mining
and Collaborative Filtering. 2021 International Conference on Science & Contemporary Technologies
(ICSCT), Science & Contemporary Technologies (ICSCT), 2021 International Conference On, 1–5.
Gong, S., Ye, H.-W., & Zhu, X. (2009). Item-based collaborative filtering recommendation using self-
organizing map. 2009 Chinese Control and Decision Conference, Control and Decision Conference,
2009. CCDC ’09. Chinese, 4029–4031.
Kamukwamba, Y., & Chunxiao, L. (2021). A Novel Algorithm Using Content-Based Filtering Tech-
nology in Apache Spark for Big Data Analysis. 2021 4th International Conference on Artificial Intel-
ligence and Big Data (ICAIBD), Artificial Intelligence and Big Data (ICAIBD), 2021 4th International
Conference On, 22–26.
Khatter, H., Arif, S., Singh, U., Mathur, S., & Jain, S. (2021). Product Recommendation System for E-
Commerce using Collaborative Filtering and Textual Clustering. 2021 Third International Conference on
Inventive Research in Computing Applications (ICIRCA), Inventive Research in Computing Applications
(ICIRCA), 2021 Third International Conference On, 612–618.
Kusumawardhani, N. K., Nasrun, M., & Setianingsih, C. (2019). Web Recommended System Library
Book Selection Using Item Based Collaborative Filtering Method. 2019 IEEE International Conference
on Engineering, Technology and Education (TALE), Engineering, Technology and Education (TALE),
2019 IEEE International Conference On, 1–8.
Lourenco, J., & Varde, A. S. (2020). Item-Based Collaborative Filtering and Association Rules for a
Baseline Recommender in E-Commerce. 2020 IEEE International Conference on Big Data (Big Data),
Big Data (Big Data), 2020 IEEE International Conference On, 4636–4645.
Padhy, S. K., Singh, A. K., & Vetrivelan, P. (2022). Item-Based Collaborative Filtering Blockchain
for Secure Movie Recommendation System. Lecture Notes in Electrical Engineering, 792, 937–948.
doi:10.1007/978-981-16-4625-6_93

191
 
Recommendation System

Riswanto, E., Robi’in, B., & Suparyanto. (2019). Mobile Recommendation System for Culinary Tourism
Destination using KNN (K-nearest neighbor). Journal of Physics: Conference Series, 1201(1), 012039.
doi:10.1088/1742-6596/1201/1/012039
Tai, Y., Sun, Z., & Yao, Z. (2021). Content-Based Recommendation Using Machine Learning. 2021 IEEE
31st International Workshop on Machine Learning for Signal Processing (MLSP), Machine Learning
for Signal Processing (MLSP), 2021 IEEE 31st International Workshop On, 1–4.
Wayan Priscila Yuni Praditya, N., Erna Permanasari, A., & Hidayah, I. (2021). Designing a tourism
recommendation system using a hybrid method (Collaborative Filtering and Content-Based Filtering).
2021 IEEE International Conference on Communication, Networks and Satellite (COMNETSAT), Com-
munication, Networks and Satellite (Comnetsat), 2021 IEEE International Conference On, 298–305.
Yadalam, T. V., Gowda, V. M., Kumar, V. S., Girish, D., & M., N. (2020). Career Recommendation
Systems using Content based Filtering. 2020 5th International Conference on Communication and
Electronics Systems (ICCES), Communication and Electronics Systems (ICCES), 2020 5th International
Conference On, 660–665.

192
 193

Chapter 10
Case Studies on the Use of
Data Mining Techniques
in Data Science

ABSTRACT
This chapter presents data science research conducted by authors Sarawut Ramjan and Jirapon Sunkpho,
showcasing the use of RapidMiner to gather data from social networks. The research includes the analysis
of customer satisfaction for mobile applications, predicting condominium prices in Bangkok, and dis-
covering demand and supply patterns based on Thai social media data using the association rule mining
approach. Additionally, the chapter touches on other topics such as corrosion under insulation severity
classification for carbon steel in a marine environment and variables that influence pilot’s safe driving.
The chapter is beneficial for readers without data science experience and for businesses interested in
employing data mining techniques. The research highlights that software application skills are not the
only important factor, but also understanding the processes of data science, such as data exploration,
data pre-processing, data mining, and data presentation, which are essential and useful skills.

INTRODUCTION

This book has mentioned data mining techniques both in nature Supervised-Learning and Unsupervised
Learning using RapidMiner as a data analysis tool. Several case studies from the author’s researches are
presented as examples for analyzing real data. They are not only the application of data mining techniques
to process the data, but also cover other processes of data science work. The case studies consist of 1.
Analysis of customer satisfaction in using Mobile Application, published in the journal “Thammasat
Review”, 2. Predicting Condominium Price in Bangkok Using Web Mining Techniques, published in the
journal “Srinakharinwirot Research and Development”, and 3. An Association Rule Mining Approach to
Discover Demand and Supply Patterns Based on Thai Social Media Data, published in an International
Journal of Knowledge and Systems Science. 4. Corrosion under Insulation Severity Classification for
Carbon Steel in Marine Environment which is a part of Independent Study of Data Science Program that

DOI: 10.4018/978-1-6684-4730-7.ch010

Copyright © 2023, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
 
Case Studies on the Use of Data Mining Techniques in Data Science

author is advisor as well as 5. Variables Influence Pilot’s Safe Driving which is a part of Independent
Study of Data Science Program that author is an advisor.

THE CASE STUDY ON ANALYZING CUSTOMER SATISFACTION

ON A MOBILE APPLICATION USING DATA MINING
TECHNIQUES (SUNKPHO & HOFMANN, 2019)

The case study shows the analysis on customer satisfaction using mobile app services of state-owned
energy enterprises in the metropolitan area of Bangkok, Thailand. The analysis follows CRSIP-DM Pro-
cess, which this book has already mentioned in Chapter 1. The researcher found that Data used to reflect
performance of the state energy enterprise is the satisfaction of users on services related to information
technology and one important service, the Mobile App.
Mobile App is available to users as they can access the Mobile App to check the amount of energy usage
each month. The app notifies important news, such as the cessation of energy services for infrastructure
repairs, the area where the power is interrupted and unable to provide service, and payment channel.

Business Understanding

In the Business Understanding process, the key is that Energy enterprises have questioned the variables
that effect customer satisfaction and dissatisfaction so that SOEs can develop features. Therefore, the
researcher has formulated a problem for data analysis, namely Classification of data that should be used
to predict customer satisfaction.

Data Understanding

After that, the researcher entered the Data Understanding process. Datasets collected customer satisfac-
tion on mobile apps were imported into RapidMiner software to conduct surveys. The data are satisfac-
tion level, gender, age, education, amount of mobile app usage per year, mobile app usage per month,
reliability and mobile app usage in emergency situations. From the survey data, it was found that there
were 81% of customers satisfied with using the mobile app at a level 4 - 5 on the Likert Scale, and 19%
of the customers showed a satisfaction level of 3 or lower.

1. Data Exploration: The researchers then used Tableau software to perform Data Exploration for
deeper data, which consists of Customer satisfaction on App Feature and User Experience. The
overall satisfaction level was at a very level. However, the least satisfied part of the data was data
security. As for the App Feature, the part which need to be improved and showed the least satisfac-
tion was the Payment Feature. In the User Experience section, the overall satisfaction was at very
satisfied level.
2. Data Quality Verification: Researchers have prepared the data by manipulating the Missing Value.
It was found that the Demographic Data had a total of 26 Missing Values. There was no missing
value in terms of reliability and usability of the Mobile App in emergency situations. There was 0
data because the customer had never used the Mobile App in such a manner.

194

Case Studies on the Use of Data Mining Techniques in Data Science

3. Correlation Analysis: The researchers analyzed the data with Correlation Analysis to analyze
the relationship of the Attribute within the data set, and found that there were data related to each
other, consisting of Group 1: Data Accuracy and Data modernization, Group 2: the attractiveness
of the User Interface and the Ease of Use which are considered the User Experience, and the 3rd
group is Notifications and Energy History which is in the App Features group.
4. Data Preparation: The missing Value data in Demographic Data was filled using the Impute
Missing Value technique within the RapidMiner software. The rest of the data within the Attribute
was used fill the empty data with the similar data. The objective of the data analysis is to find what
variables cause customer satisfaction and dissatisfaction. The researchers revised the Likert Scale
data to form a binary class with only two values. Changes were made to the format of the data. For
satisfaction data on a 4 - 5 scale, it was changed to satisfactory, and the level of satisfaction from
3 and lower was changed to dissatisfactory.
5. Modeling: The researchers used 3 data mining techniques to analyze the data: Naïve Bayes,
Decision Tree and Logistic Regression, and divided the data into 2 parts according to the principle
of Supervised-Learning. 80 percent was for teaching machine learning and 20 percent was for model
performance testing. The researchers then created a model with all 3 data mining techniques. The
data sets were imported into the model. When testing the analysis accuracy, the accuracy level
of Naïve Bayes, the Decision and the Logistic Regression were 92.76%, 91.03% and 92.41%,
respectively.
6. Evaluation: Analysis of the data showed that Naïve Bayes was slightly more accurate than Decision
Tree and Logistic Regression. However, when considering Precision and Recall, the Decision Tree
was the most accurate. Therefore, the researchers chose to consider the results of the Decision Tree
data analysis, and it revealed that the most important variables that caused customer satisfaction
and dissatisfaction were Ease of Use of Mobile App, and Payment Feature.
7. Deployment: After analyzing the data, the researcher advised energy enterprises to develop a
new version of the mobile app, which should include improvements in ease of use and payment
features in order to increase customer satisfaction and reduce dissatisfaction. After the enterprise
has developed a new version of Mobile App and used for customers in the metropolitan area, the
researcher will collect the data and analyze the results in a similar manner to test the performance
of the new version of Mobile App to examine.

The case studies from the research paper “Analyzing Customer Satisfaction of a Mobile Application
using Data Mining Techniques” can be used as a guide for data scientists who study data mining tech-
niques and practice using RapidMiner software. The scientists are required to understand not only about
the methods of data analysis and the use of the software, but they also need to understand the processes
under the CRSIP-DM, so that they can prepare data and analyze results.

THE CASE STUDY ON PREDICTING CONDOMINIUM PRICE IN BANGKOK

USING WEB MINING TECHNIQUES (SUNKPHO & RAMJAN, 2020)

The sales rate of condominiums in Bangkok, Thailand among the Generation Y and among condominium
speculators is very high. This group of condominium buyers use variables to evaluate the appropriate
condominium price, such as the distance to the sky train or subway. These variables are used to determine

195
 
Case Studies on the Use of Data Mining Techniques in Data Science

the price the buyer thinks appropriate for the purchase and to determine the selling price. The environ-
ment of the city where the condominium project is being built is another important variable.
Therefore, the researcher focused on analyzing variables suitable for forecasting condominium prices
so that the results are accurate and appropriate to the characteristics of the city in which the condominium
project is being built. The researcher therefore defined the control variables as a study of the variables
related to the pricing of condominiums in Bangkok, Thailand. The research process is as follows.

Data Collection Using Web Crawling

The researchers used Web Crawling, one of Web Mining’s techniques, to extract data on the condominium
purchases from the website, hipflat.co.th which covers more variables than other similar websites. The
data set was brought to teach machine learning under the concept of Supervised-Learning. The research-
ers obtained data on 1,465 condominiums and within the dataset there were a total of 15 attributes.

• Project location
• List of sky train stations near the condominium
• List of subway stations near the condominium
• The average distance from the condominium to the surrounding sky train stations
• Average distance from the condominium to the surrounding subway stations
• Name of the condominium project
• Number of floors of each project
• Number of rooms in each project
• Year the construction was completed
• Names of project owners
• Price per square meter
• The price rate that has changed since the end of the last year
• The price rate that has changed since the end of the last quarter
• Rental rate that has changed since the end of the last year
• Return on Investment

Data Exploration

The researcher used Correlation Analysis to analyze the variables related to pricing and found that the
following 4 attributes affect the condominium pricing; average distance from condominium to sur-
rounding sky train station, the average distance to the surrounding nearest subway station, return on
investment, and the rental rate change since the end of last year. The number of floors and the year of
completion positively influences on the price per square meter. However, the average distance from the
condominium to the surrounding sky train stations, and the average distance from the condominium to
the surrounding subway stations negatively influence on the price per square meter.

Data Preparation

The web-crawling dataset contained a number of missing values, so the researchers needed to manipu-
late the data by filling in the mean into the attribute where there was the missing value, consisting of

196

Case Studies on the Use of Data Mining Techniques in Data Science

the average distance from the condominium to the surrounding sky train stations, the average distance
from condominium to surrounding subway stations, the number of floors and the number of rooms in
each project.

Data Modeling

The researchers used five different data mining techniques to forecast condominium prices: Generalized
Linear Model, Deep Learning, Decision Tree, Random Forest, and Gradient Boosted Trees. RapidMiner
was used as a modeling and data analysis tool. The researchers then used the split test technique to di-
vide the data into two parts: 80 percent for teaching machines learning, and 20% for testing the model’s
performance.
After testing the model’s efficiency in forecasting five condominium prices, the researchers found
that deep learning data mining technique is the most accurate in forecasting. The data scientist adjusted
the model to increase efficiency by adding a Discretized Process model within the RapidMiner software
to the model to classify the data of the Attribute. The average distance from the condominium to the sur-
rounding sky train stations and the average distance from the condominium to the surrounding subway
stations were divided into 4 intervals. These included distances less than 500 meters, distances between
500 and 1000 meters, distances between 1 and 2 kilometers, and distances greater than 2 kilometers. The
researchers then used the Distance Measure model within the RapidMiner software to remove outlier
data. Regarding the Attribute price per square meter, the researchers found that after adjusting both parts
of the model, the model had a higher accuracy in forecasting condominium prices.
In the last step, the researcher transferred data from Attributes; including Return on Investment, Rates
that have changed since the end of last year, the price change since the end of the last quarter, Rental
rates changed since the end of last year, to teach the machine. On the contrary, it was found that after
adjusting the data, the accuracy of condominium price forecasting declined.

The Research Result

Therefore, the researchers concluded that Important variables to consider when determining condominium
prices. The first variable is the year of construction completion. The new condominiums have higher
price per meter. Secondly, the more numbers of floors in the project, the higher the condominium price
is. On the other hand, if the average distance from the condominium to the sky train station and the
subway station is very far, the price of the condominium will reduce. In addition, the more numbers of
rooms per project, the lower the price of condominiums. The research also found that if the names of
the project owners are listed in the stock market, the prices of condominiums is higher than the condo-
miniums whose owners’ names are not listed in the stock market.
The results from this research are also consistent with other researches, such as condominium price
forecasting in Japan (Diewert & Shimizu, 2016) and Singapore (Lim, et. al, 2016), which analyzed the
variables used to determine condominium prices.

197
 
Case Studies on the Use of Data Mining Techniques in Data Science

THE CASE STUDY ON ASSOCIATION RULE MINING APPROACH

TO DISCOVER DEMAND AND SUPPLY PATTERNS BASED ON
THAI SOCIAL MEDIA DATA (TANANTONG & RAMJAN, 2021)

Twitter has become a digital marketing channel for online sellers by increasing the number of Retweets
on the products and services. To create a Tweet, online sellers also use Hashtags to categorize customer
interests such as #KoreanArtistConcertTickets to make the Tweet appear to customers who are interested
in the topic. And in many cases, the Hashtag ranks at the top which means it is the viral, such as politi-
cal phenomena. Online sellers also use hashtags on the subject, so that the Tweets they generate appear
alongside other Tweets of the interest of society. Therefore, the researchers used Association Rule Min-
ing to analyze the correlation rules of demand and sales from a dataset of Tweets, Retweets, Likes and
Hashtags related to goods and services on Twitter. The research process is as follows:

Data Collection

The researchers used the Twitter API along with Python to extract trading on Twitter, where online sellers
and buyers interact for the 24 hours. It extracted only the buy and sell data with a total of 24,464 instances.

Data Cleansing

To prepare data for processing with affinity rules, the researchers cleaned up the data by examining the
duplicates and removing them from the dataset, and then checked the tweets without comments, which
means the products did not have response from customers. Researchers then deleted the data that had
such characteristics from the data. In the final step, the researchers considered only the attributes involved
in the analysis of the data, and removed the irrelevant attributes such as profile_image_url_https, which
stored information about Twitter users, until there were only 12 attributes, with a total of 7,475 instances.

Data Labelling

Among all 12 Attributes associated with trading on Twitter, the researcher asked 3 Knowledge Experts
for their opinions on which attributes to be analyzed with correlation rules. The experts recommended
categorizing data into 5 groups: Demand Tweets, Supply Tweets, Demand and Supply Tweets, Comment
Tweets and Other Tweets.

Hashtag Extraction

The researchers extracted 3 hashtags from the data: Demand Tweets, Supply Tweets, and Other Tweets.

Association Rule Mining

The researchers created all possible Frequent Item Sets from the dataset with the Apriori algorithm dis-
cussed in Chapter 8. The Support and Confidence Thresholds were set for Demand at 0.02, for Supply at
0.01, and for Other Tweets at 0.0001. The researchers then analyzed the data in order to create 1-itemset,
2-itemset, and 3-itemset correlation rules, and selected 2 strong correlation rules: Right-Hand-Side (RHS)

198

Case Studies on the Use of Data Mining Techniques in Data Science

which shows Demand, Supply, and Lift-Hand-Side (LHS) which doesn’t show the Hashtag type. The
correlation rule displays the Hashtag in LHS format along with the RHS format.

The Experimental Result

The researchers found that online sellers should offer their products and services in conjunction with
the use of hashtags related to the social phenomenon or trends in order to attract the shoppers’ interest.
During the time of the experiment, there were Korea super stars to hold concerts in Thailand. Hashtags
associated with the super stars became so popular that online sellers could present their products and
services using the Hashtags. Also, the most important hashtags are buying and selling. This should be
presented within the Tweet so that customers can understand and access the products or services the
seller wants.

CORROSION UNDER INSULATION SEVERITY CLASSIFICATION FOR CARON

STEEL IN MARINE ENVIRONMENT (THAMMASIRI & RAMJAN, 2021)

In the oil transportation industry Oil conveying equipment is made of insulated steel. Such insulators
are meant to maintain the temperature of the oil transported within the pipeline. Engineers need to open
the insulation to check for corrosion by continuously expanding the opening until all erosion points are
found. The resulting corrosion could lead to an undersea oil spill that could contaminate the sea. destroy
marine ecosystems or even cause fires that destroy nature and human life. The entire process runs on
an offshore oil station, making it a difficult process for pipeline repairs and huge maintenance costs.
Data scientists need to take a dataset of rust corrosion on oil pipelines and analyze them to classify
where rust may occur so that engineers can open the insulation at the location of the corroded steel from
the rust. Data analysis takes existing data sets to teach machine learning and gives machines the ability to
classify the degree of corrosion of steel so engineers can diagnose preliminary milling situations before
deciding to open the insulator covering the duct for a real local inspection. The study and analysis of
this data can help reduce the time for pipeline repairs and reduce the cost of maintenance of the pipeline.
The research used the environment to study and analyze data from an oil production platform in the
Gulf of Thailand of an oil company from Thailand. Resources for collecting datasets used to study and
lead to modeling in steel corrosion prediction include Pipeline Corrosion Inspection Report, Pipeline
Oil Temperature Report and Plant. The researchers then took the collected data and compiled it into a
dataset that could be used for analysis by data mining techniques. Data exploration process, the research-
ers surveyed the data to determine how appropriate the patterns of the datasets were to be analyzed us-
ing data mining techniques and to the research question. Researchers create a visualization for the data
exploration consisting of bar charts, scatter plot and box plot.
Next, in the data pro-processing process, the researchers dealt with the missing value by fulfil the
incomplete data, and feature engineering stage, researcher prepare a standardization of data that came
from different measurement by label encoding, one hot encoding and max-min normalization. During the
tuning of that data, researchers deal with the imbalanced data using a SMOTE algorithm. The research-
ers then divided the dataset into training set to teach machine learning. and testing set to approve the
performance of the classification model. In addition, this step researchers fine-tuned the hyperparameters
with the GridSeachCV technique.

199
 
Case Studies on the Use of Data Mining Techniques in Data Science

This experiment aims to explore classification techniques from different data mining methods that
most appropriate to the data set and research questions. Classification techniques are explored through
a literature review that attempts to explore and consolidate the theories and research models involved in
this research to compile possible solutions that can be applied to classify the degree of erosion of cor-
rosives of oil pipeline. After literature review process, this research discovered a possible classification
technique including decision tree, random forest, XGBoost and multilayer perceptron. The researchers
compared the classification accuracy of each technique to conclude which technique is best suited for
data analysis under research environments.
Experimental data set consists of 18 features and 434 instances. Features inside a data set includ-
ing the plant where the pipeline was manufactured, the year the pipe was installed, the size of the pipe,
the fluid inside the pipe, the year the engineer inspected it for corrosion, the equipment was inspected,
the condition of the jacket, and the condition of the silicone, jacket damage scale and silicone, type of
insulation, the minimum average temperature of the internal fluid, The maximum mean temperature of
the internal fluid, the average temperature of the fluid inside, the altitude of the device location from
sea level, location of the device in the Y-axis, location of the device in the X-axis, and the degree of
corrosion severity which is a label. The data classification is divided into two sections: insulation dam-
age assessment and pipe corrosion assessment. Insulator damage is classified into 4 classes: no damage,
damage is less than 1 inch in circumference, damage is 1 inch in circumference, damage is larger than
1 inch. Corrosion in pipes is divided into 3 classes: non-corrosive, mild and severe. During the feature
engineering process, the researchers created new features that could affect a label classification consists
of paint age, the difference between the highest and lowest temperatures, the risk of pipe corrosion and
the rate of corrosion.
From the experiment, From the experiment, this research employed a confusion matrix to be a tool
for evaluate value an accuracy of each model experiment. It was found that decision tree model has an
accuracy value of data classification at 71.8%, multilayer perceptron model at 72.9%, random forest
model at 81.3% and XGBoost model at 82.2%. While testing for classification accuracy with machine
learning is as reliable as data science theory, this research needs to prove the accuracy of real-world
experiments. The researchers deployed the XGBoost classifier model to a web form to allow engineers
to input parameters and then have the model classify rust corrosion levels for decision-making on oil
pipeline maintenance. From real-world testing, it was found that the model has 75% accuracy in clas-
sifying corrosion data for pipelines, based on testing by engineers working in pipeline repairs.
After data analysis and research studies, the researchers found that the limitations that could be de-
veloped in data science research studies in other future research and study environments similar to this
one. The data sources for this research were compiled from quantitative frequencies stored in a single
sensor technology of each oil platform. If future research can collect frequency data from more than
one sensor technology in each oil platform, the results of the experiments could be more accurate. In
addition, in the case of other research It can collect data from different environments, such as deserts,
giving machine learning the opportunity to learn features of the data goes beyond this research and may
make it more accurate to classify the data.

200

Case Studies on the Use of Data Mining Techniques in Data Science

VARIABLES INFLUENCING PILOT’S SAFE DRIVING

(MANEECHAEYE & RAMJAN, 2022)

While flying is considered the safest journey out of all means, accidents can occur due to several variables.
One of the interesting variables is accidents caused by pilot failure. Strict adherence to aviation standards
is one of the ways to prevent human error. This research aims to study the variables that influence the
safe driving behavior of pilots at the organizational, squadron and individual levels. In this research
study, the researchers relied on domain expert to collect qualitative data to study the overall condition
of aviation safety in Thailand and pilot behavior that does not meet aviation standards. The researchers
then applied the recommendations from domain experts along with the literature review data to develop
a questionnaire tool, and the result of collecting data is 610 instances.
After data collected, during data exploration process, researchers analyzed a central tendency of each
feature to explore a data distribution. This research relies on exploratory factor analysis techniques to
group 15 features into 4 factors consists of safety climate, knowledge of aviation safety standards, the
motivation to operate safely and last factor is the pilot’s compliance with aviation standards which is a
label. To analyze the variables that influence the pilot’s safe driving behavior, the researchers used a data
mining technique which is a multiple regression to analyze a data set. Data set is split to training set at
75% to train a machine learning as well as testing set at 25% to test an accuracy of prediction model. This
research evaluates an accuracy of the data mining model and found that mean absolute error (MAE) at
0.47, mean square error (MSE) at 0.42, Root mean square error (RMSE) at 0.65 and R-squared at 0.62.
The results of the data analysis revealed that the safety climate had a significant effect on pilots’
compliance with aviation standards. Knowledge of aviation safety standards has a significant effect on
the pilot’s compliance with the aviation standards. The motivation to operate safely has a significant
effect on the pilot’s compliance with aviation standards.

CASE STUDY

The Design and Implementation of Attraction Guidance

Systems Using Data Mining Techniques in Conjunction with
Collaborative Filtering Technique (Nan et al., 2022)

Designing routes and attractions in corresponding to the personalization of tourists is an important

process for tourism businesses. Understanding the conditions for each targeted tourist is critical to the
design of routes and attractions. There are many ways to analyze data, and one of the methods that can
be applied is the data scientist’s approach to analyzing what customers are likely to be interested in for
each product and service. Such an approach can be used for analyzing the sights that tourists, custom-
ers, like and dislike.
Applying Collaborative Filtering for tourist attraction analysis for, the results obtained from the analysis
will help reduce the problems that can arise for tourists while traveling in various places. In 2022, there
was a research paper related to the design of tourist guide systems using Collaborative Filtering (Nan,
Kanato, & X, 2022). Initially, this research collects data from each traveler’s search for attractions on a
web browser, along with traveler opinions about tourist attractions on social media. The data were ana-
lyzed in order to recommend suitable attractions to tourists in the same market segmentation. However,

201
 
Case Studies on the Use of Data Mining Techniques in Data Science

once the Data Exploration process was completed, the researchers selected the features to be used in the
data analysis with collaborative Filtering consisting of the current attractions that tourists are studying
and the ratings that tourists have given for each tourist destination they have been to. The researchers
then manipulated the missing value data and experimented it with Collaborative Filtering to increase
the efficiency of data analysis. This research also brings other data for further experiments, including
travel expenses, location of tourist attractions, the distance to travel to each attraction and travel time to
each attraction to ensure the accuracy of suitable recommendation.
This research also suggests that in the future, other researchers can use real-time coordinates as well
as real-time weather data to help make an appropriate recommendation about attractions.

Guidelines for Developing a System to Provide a Recommendation

on Streaming Online with Data Mining (Liao et al., 2021)

Today, with the growth of digital utilities that are stable and supportive to high-speed loading large data,
the general public can access to the Internet faster than in the past. Businesses have come up with the idea
of Live stream to present their products as they can assess the customer’s attitude towards the product
demonstrations. Customers can place their desired products during the Live. Facebook was one of the
first to use Live Feature to showcase or demonstrate their products. Customers who are also Facebook
users can then view the product display in real-time or watch the demonstration that was shown earlier.
They can also interact with sellers and place their order via Chat. The seller can also show the nature of
the product, how it works, product prices, pros and cons of using products that are appropriate for each
customer’s characteristics through Facebook Live. While businesses recognize the value of offering
goods and services through online streaming, businesses face challenges of collecting and analyzing
customer data through customer interaction because such interactions occur only for a short time and
cannot reflect long-term consumer behavior of the product.
Other e-Commerce channels, such as e-Retailer, can provide a recommendation system for customers
who have experienced shopping through the website in the past. And when the customers return to make
repeat purchases, the website can recommend items that customers are likely to be interested or close
to the products they’ve used. The website can even analyze customer’s market segmentation and make
recommendations that are close to what customers in the same segment are interested in as well. So, the
ability to analyze data in a big data will be an important variable for a website. The more information a
website has on its customers, the better. The website will be accurate to analyze and recommend. The
more items that correspond to the needs of the customer, the more this process incentivizes customers
to choose products and make purchases instead of simply visiting the website and turning off the web
browser. So, understanding and knowing the characteristics of the customer is an important condition
for the recommendation system.
Choosing the right Live Platform for each business’s customers is essential for selling products through
Live stream. In 2021, there was research on ways to develop online streaming advisory systems (Liao,
Widowati & Chang, 2021) as follows.

• Customer basic information

• Customer’s online shopping habits
• Motivation to shop through customers’ online channels
• Motivation to interact with sellers via Live Streaming

202

Case Studies on the Use of Data Mining Techniques in Data Science

• Customer behavior towards the merchant’s live streaming

• Variables that affect decision to purchase online
• Payment methods
• Customer behavior while watching Live streaming
• Customer behavior when interacting with sellers via Live streaming
• Reasons to watch Live streaming
• Live Stream Platform that customers choose to watch
• Types of media content that customers often watch live stream, such as politics, entertainment,
etc.
• Motivation to use Stream Platform, for example, to kill time or reduce stress
• Marketing Promotion
• Activities that occur while watching Live; donations or interactions with vendors
• Marketing promotion that affects the purchase of goods, such as offering new products to customers
• Recommending items suitable for customer’s needs; brand or payment method

According to the analysis of the data, it was found that customers enjoy interacting with sellers via
Live too. Therefore, both the seller and the platform type affect the customer’s interest. Customers
also recommend their favorite streaming to their peers around them. When customers pay attention to
a streamed product, they often share that content on their platform. Customers also like reading other
customers’ comments on Streaming. This is a hidden reason for customers changing to online shopping
instead. The reputation of the Live and the overall beauty of the person who live affects the purchase
decision of the customer. In addition, businesses should pay attention to the production of content that
is consistent with the current popular issues and pay attention to the quality of streaming. The price
reduction during live is also another variable that affects the purchasing decision of customers.
Merchants can also use Streaming to offer new products or special prices to VIP customers. Men-
tioning regular customers in Streaming also affect the speed of commenting by existing customers and
increase interaction with other customers. This is similar to referral Item used by previous customers
and affects other customers’ purchasing decisions to buy this referred Item.
According to all case studies, when data scientists intend to develop a referral system for products
or services, understanding the datasets in the organization or outside the organization alone may not
be enough. It is necessary to consult a Domain Expert or even collect data to study the variables and
factors that affect the conditions for providing advice on products and services that are suitable for each
group of consumers so that the items received from the recommendation system can really affect the
purchasing decision.

CONCLUSION

According to the case studies on the author’s researches, data scientists need to study data mining
techniques and applications of RapidMiner software, along with doing experiment throughout the data
science process. It starts with collecting and extracting data from the source, surveying, pre-processing,
choosing a data mining technique correspondent to the questionnaire, and finally presenting the results.
Therefore, rather than focusing only on data analysis techniques from the beginning, the scientists should

203
 
Case Studies on the Use of Data Mining Techniques in Data Science

analyze the questions thoroughly so that other relevant processes are aligned, and ultimately led to ef-
ficient analysis.

EXERCISE QUESTIONS

1. Determine the topic used to analyze the problems.

2. From No. 1, find the Dataset on the Internet that can be used for analyzing the problems.
3. From No. 2, outline the principles and rationale for analyzing the problems, and indicate the data
mining techniques used in problem analysis.
4. From No. 3, use RapidMiner software to process Data Exploration and Data Pre-Processing. The
data mining techniques are then used to analyze the data.
5. From No. 4, write a report showing the results of data analysis and conclusions.

REFERENCES

Liao, S.-H., Widowati, R., & Chang, H.-Y. (2021). A data mining approach for developing online stream-
ing recommendations. Applied Artificial Intelligence, 35(15), 2204–2227. doi:10.1080/08839514.202
1.1997211
Maneechaeye, P., & Ramjan, S. (2022). Components and Indicators Development with Guidelines to
Encourage Safety Behavior among Flight Crews in Thailand (Independent Study). Thammasat University.
Nan, X. (2022). Design and Implementation of a Personalized Tourism Recommendation System Based
on the Data Mining and Collaborative Filtering Algorithm. Computational Intelligence and Neurosci-
ence, 1424097. PMID:36093493
Sunkpho, J., & Hofmann, M. (2019). Analyzing Customer Satisfaction of a Mobile Application using
Data Mining Techniques. Research Administration Division, Thammasat University.
Sunkpho, J., & Ramjan, S. (2020). Predicting Condominium Price in Bangkok using Web Mining Tech-
nique. Srinakharinwirot Research and Development (Journal of Humanities and Social Sciences), 12(24).
Tanantong, T., & Ramjan, S. (2021). An Association Rule Mining Approach to Discover Demand and
Supply Patterns Based on Thai Social Media Data. International Journal of Knowledge and Systems
Science, 12(2), 1–16. doi:10.4018/IJKSS.2021040101
Thammasisir, S., & Ramjan, S. (2021). Corrosion under Insulation severity Prediction for Carbon Steel
in Marine Environment using Data Science Methodology (Independent Study). Thammasat University.

204
 205

Chapter 11
Data Mining for Junior
Data Scientists:
Basic Python Programming

ABSTRACT
The availability of off-the-shelf tools for data mining has made it easier to process data. However, in many
cases, such software packages are not flexible enough to allow for algorithmic improvements. Therefore,
data scientists need to write computer programs to customize the processing methods in tandem with
the software packages. This chapter introduces Python, a programming language that provides libraries
that support data science work. The content includes Python programming syntax, such as variables,
structured programming, decision-making programming, recursive programming, data structure han-
dling, and file handling. Additionally, the chapter introduces Google Colab as a tool for programming
experiments. It provides a crucial foundation for data science students who are going to process data
using data mining techniques in the next chapter using Python programming. Although data scientists
don’t need a deep understanding of computer programming, learning computer languages is essential,
and this chapter caters to beginners.

INTRODUCTION

A number of industries, nowadays, focus on data science as they recognize the value of data to support
decision-making (Raschka & Mirajalili, 2018). Whether it is the medical industry, automotive industry
or the scientific community, data scientists have become a sought-after position for many organizations.
A data scientist can either be developed from the knowledge expert within the organization itself or be-
ing adopted from outsiders. Anyhow, every organization analyzes data on the principle of questioning,
seeking resources, exploring data and pre-Processing, analyzing data using data mining techniques, and
presenting the results to the questioners. At present, there are many tools for exploring and preparing
data, as well as machines for using data mining techniques to analyze data. One of these tools is Python,
which provides a library of commands for developing visual data, exploring and preparing data, as well

DOI: 10.4018/978-1-6684-4730-7.ch011

Copyright © 2023, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
 
Data Mining for Junior Data Scientists

as analyzing data with data mining techniques. In addition, data scientists can process data with Python’s
algorithms without limiting the size of the data. Python itself is considered an open source that allows
organizations to analyze data with data mining techniques, which reduces the cost of digital resources
(Sneeringer, 2016: Reges, Obourn & Stepp, 2019). Therefore, Python is a useful tool for beginners in
data science (Bowles, 2015: Joshi, 2017).

INTRODUCTION TO PYTHON

Python, developed by Buido van Rossum in 1980, is a high-level language which is so close to human
language, making it ideal for beginners in computer programming languages (Sunkpho & Ramjan, 2020:
Tanantong & Ramjan, 2022). It also has an easy-to-remember syntax, so software developers can use
Python for programming in both the Structural Programming and the Object-Oriented programming
styles (Hill, 2015: Bader, 2018: Mueller, 2018: Lubanovic, 2019). Python is an open software source,
so software developers worldwide can work together to develop Python without the cost of licensing
(Stewart, 2017: Zelle, 2010).
Software developers can use software editors such as Google Colab to write Python and use the in-
terpreter to translate Python in order to run the digital device with the process as shown in the following:

1. Software developers write a set of commands on a software editor.

2. The interpreter translates Python into the language that the digital devices can operate.
3. The digital devices accept the commands for processing.
4. The digital devices operate as commanded.
5. Software developers check the operation accuracy.

In the figure, software developers can write a set of commands on the Software Editor and then uses
Interpreter to translate Python into the format that can be run by digital devices. The digital devices
then receive the command and process it to perform the tasks as designed by the software developers. In
the final step, software developers can verify whether the digital devices work as designed so that their
command sets can be improved to make digital devices work more accurately.
Python is Case Sensitive, meaning lowercase and uppercase English letters are different. For example,
dX and DX, when interpreted, the letters will totally be different. Although developers have different
duties from data scientists, when data scientists intend to use programming for data analysis, they can
begin their Python study with the following basic commands.

Variables

Data scientists can analyze data by packing it into variables within Python. The variables act as a con-
tainer containing data for data transfer, processing, and display. The process goes as presented in the
figure below.

206

Data Mining for Junior Data Scientists

Figure 1. Operation of variables

Data scientists can determine the data containing variables. In the example, the data scientists assign
all 3 variables. Variable 1 stores the first data which is number 8; and variable 2 stores the second which
is number 5. The numbers in variable 1 and variable 2 are then combined, so the result is 13, which is
then included in the third variable. Thus, data scientists can define variables to contain the data and write
Python programs based on the algorithm designed for each data mining technique.
To determine variables, letters, numbers, or underscores can be used, but it cannot begin with the
number like dX or StudentCode. However, variable determination should be easy for data scientists to
recall. Therefore, data scientists can begin programming in Python with the following syntax.

variable = “alphabetic data”

variable = numeric data

From the syntax, data scientists can pack alphabetic and numeric data into variables and display the
data through “print” command as presented in the example below.
In the figure, the variable name contains the alphabetic data “Jirapon Sunkpho” and variable age
contains the numeric data which is 50. The data within the variable are then output via “print” com-
mand. After the data scientists have filled the data into the variable, they can also perform mathematical
operations on those variables to process the data as in this grammatical syntax.

variable = variable arithmetic operator variable

From the syntax, data scientists can pack numerical data into variables. The variables are then
performed with mathematical operations to obtain the result. Then the result is displayed via the print
command as shown in Figure 4.

207
 
Data Mining for Junior Data Scientists

Figure 2. Example of variable determination

Figure 3. An example of using variables to perform arithmetic operation

208

Data Mining for Junior Data Scientists

In this example, variable num1 contains 5, variable num2 contains 8. When the numbers of num1 and
num2 are added together, it equals 13. It is then put into num3 and output through print command. Ac-
cording to the definition of alphabetic and numeric variables, within Python the types of variables used
in computation are defined as float which represents a real number such as 0.5, and int which represents
an integer such as 5. Data scientists can examine the type of data within a variable as in the syntax.

print(type(variable))

From the syntax, data scientists can use type command to check the data type within the variable by
working in accordance with the print statement, as shown below.

Figure 4. Types of variables

In the figure, variable ex contains number 6, and the variable ed contains number 9.0. When using
print(type()) command to display the data type, the result is an integer, which means variable ex con-
tains integer-like data. And when displaying the data in variable ed, it is number 9.0 which is a decimal
number. When data scientists intend to change the variable type for data processing, the programs can
be written as in the following syntax.

print(int(variable))
print(float(variable))

From the syntax, data scientists can use int command to convert float variables into int, and use float
command to convert int variable into float, as shown below.

209
 
Data Mining for Junior Data Scientists

Figure 5. Converting the data type

In the figure, variable ed contains number 9.0, which is float or real number, and variable ex contains
6, which is int or an integer. Data scientists can convert the data type; for variable ed, print(int(ed))
command is used to convert the data type from float to int, changing the result is from 9.0 to 9, and for
variable ex, print(float(ex)) command is used to change the data type from int to float, changing the
result from 6 to 6.0.

Mathematics Operation

Regarding the operation of variables with numbers within Python, mathematical operators are provided
for data scientists to use in conjunction with the algorithms of data mining techniques.

Table 1. Mathematical operators in Python

Operator Meaning Example

+ plus a+b
- minus a-b
* multiply a*b
/ divide a/b
% Modulus a%b
** exponentiation a**b

210

Data Mining for Junior Data Scientists

Within Python, there is a sequence of interpretations of math operators, and the Interpreter performs
the calculation of multiply, divide, plus and minus, respectively. Therefore, when data scientists design
computational algorithms for each model, it cannot decipher the formula in usual sequence, being from
left to right. Instead, it is important to consider the sequence in which the interpreter translates the lan-
guage in order to obtain the correct data analysis results. The example of data sequencing is as follows.

Figure 6. An example of a mathematical operation

In the example, variable a contains number 5, variable b contains number 3, variable c contains num-
ber 4, variable d contains number 6, and variable e contains number 2. The mathematical calculations
are then performed starting with multiplication, division, addition and subtraction respectively until the
result is -4.0. Meanwhile, the data scientists can sort mathematical operations with parentheses to arrange
the operations in addition, subtraction, multiplication, and division in sequence until the result is 12.0.
In addition to performing mathematical operations with variables, Python also provides other types
of operations to reduce programming time.
From the table, data scientists can change the configuration format to reduce programming time. For
example, c = c + 3 means taking the value inside variable c to add with the number 3 and then putting it
back into variable c. Data scientists can write programs in the form c=+3 in order to obtain the results
of the processing as in the example.
In the example, variable h and g have a starting value of 0. Then h = h + 3, so h is set as 3. The
writing style is then converted to g +=3, and the same result is obtained; g equals 3. Data scientists can
change their programming style in such a way to reduce programming time.

211
 
Data Mining for Junior Data Scientists

Table 2. Syntax and configuration examples within Python variables

Operators Used to Assign Writing a Set of Commands Without Using Operators for
Example
Variable Values Configuration
= c=3 c=3
+= c += 3 c=c+3
-= c -= 3 c=c-3
*= c *= 3 c=c*3
/= c /= 3 c=c/3
%= c %= 3 c=c%3
//= c //= 3 c = c // 3
**= c **= 3 c = c ** 3
&= c &= 3 c=c&3
|= c |= 3 c=c|3
^= c ^= 3 c=c^3
>>= c >>= 3 c = c >> 3
<<= c <<= 3 c = c << 3

Figure 7. Converting the configuration style

Logic Operation

Data scientists can process data within variables with logical data types or True and False data types
according to the following processing conditions.

212

Data Mining for Junior Data Scientists

Table 3. Logical operators within Python

Logical Operators Meaning Example

True and True are resulted as True.
and a and b
Otherwise, it is false in all cases.
False or False are resulted as False
or a or b
Otherwise, it is true in all cases.
not True and False values within a variable are transformed into the opposite values. not a

From the table, when data scientists intend to process Boolean data, they can write a program as follows.

Figure 8. Logical data processing

As shown in the figure, logical value of variables a and b are assigned as True. The logical operator
and, variables a and b are obtained. And then the data is put into variable c, and the result is presented as
True via the print command, where True and True is always true, otherwise it is false in all cases. When
variables a and b are executed with logical operator or, and the data are put into variable d, the result is
True because False or False is False; otherwise, True in all cases. And in the last command, not a has
changed the value True into False. It is then put into the variable f and output through the print command.

213
 
Data Mining for Junior Data Scientists

Comparation Operation

Whenever data scientists intend to compare the values contained within variables, comparative opera-
tions can be used. If the result of the comparison is true, it will get True value, and if the result is false,
it will get False value as presented in the table.

Table 4. Comparative operators

Operators Meaning Example

The greater-than sign is used to compare values within variables. The result is True if the value in the first
> a>b
variable is greater than the value in the second variable; otherwise, it is False in all cases.
The less-than sign is used to compare values within variables. The result is True if the value in the first
< a<b
variable is less than the value in the second variable; otherwise, it is False in all cases.
The equals sign will get the result True when the values inside both variables are equal; otherwise, it is False
== a == b
in all cases.
The not equals sign will get the result True when the values inside the two variables are not equal; otherwise, it
!= a != b
is False in all cases.
Either the greater-than sign or equal sign is used to compare values within variables. The result is True if the
>= value in the first variable is greater than or equal to the value in the second variable; otherwise, it is False in all a >= b
cases.
Either the less-than or equal sign is used to compare values within variables. The result is True if the value in
<= a <= b
the first variable is less than or equal to the value in the second variable; otherwise, it is False in all cases.

From the tables, data scientists can compare values within variables by programming as presented
in the example.

Figure 9. An example of using comparison operators

214

Data Mining for Junior Data Scientists

In the example, variable a contains number 20 and variable b contains 10. If a is greater than b, the
result is True because the value in variable a is greater than the value in variable b. When a is less than b,
the result is obtained as False because the value in variable a is not greater than the value in variable b.

CONDITIONAL OPERATION

When data scientists intend to use logical and comparative operators to command a digital device in
order to perform a particular action or repeat it, they can use the following control commands.

1. if Condition

When data scientists run a digital device in cases of logic and comparison operations, they can be
program it according to the following syntax.

if condition:
command

From the syntax, an operation takes place when the result of a condition is True. The data scientists
can write a program as follows.

Figure 10. Using if command

215
 
Data Mining for Junior Data Scientists

In the figure, variable a contains number 10, and if command is then used to compare whether the
value in variable a is equal to 10 or not. When executing the command set and the result is True, the
command after the colon is then executed. It is to display the message “Digital Transformation” through
the print command.
In case data scientists wants run a digital device when the result is false, it can be programmed with
the following syntax.

if condition:
command
else:
command

From the syntax, when the result is True, the command set will execute the command after the colon.
And if it is False, the command set will execute the command after the else: command. Data scientists
can write a program as follows:

Figure 11. Operation in case of getting result false

In the figure, in case the condition result is False, data scientists can run a digital device by writing
a set of command after the else command. As in the example, variable a has a value of 9. When the
command is equal to 10, the result is False. The device is run after the else command, and the word
“College of Innovation” is then displayed through the print command.
In case that data scientists intend to create conditions for commanding digital devices to work with
more complexity, they can write a program using the following syntax.

216

Data Mining for Junior Data Scientists

if condition:
command
elif condition:
command
elif condition:
command
else:
command

Based on the syntax, data scientists can design multiple choices of commands. If the result of the 1st
condition is True, the first set command set will be executed. On the other hand, if the result is False, the
command set will compare the second condition, and if it is True, the second statement is executed. If
the result is False, the instruction set will then compares the 3rd condition, and if True, the 3rd statement
is executed. And if False, the last statement is executed. Data scientists can write a program as follows.

Figure 12. Example of using complex conditions

In this example, the data scientists determine that there are a total of 4 alternative commands, initial-
izing variable a to 5. Condition 1 checks whether the value in variable a is greater than 10. When receiving
the result as False, it jumps to check the 2nd condition whether the value in variable a is greater than 8
or not. When the result is False, it jumps to check the 3nd condition whether the value in variable a is
greater than 6. And when the result is False again, it executes the command after else: displaying the
word “Design Thinking”.

217
 
Data Mining for Junior Data Scientists

2. Loop

When data scientists expect iterative operations, it can be either of loop statements within Python:
the for statement or the while statement.

a. For Loop

The for loop syntax can be divided into two; for Loop 1 is only when within a specified range, and
for Loop 2 represents the values contained within the variable according to the following syntax.

for Loop 1: Using the for command along with specifying the range
for i in range(n):
command

From the syntax, the scientists are required to assign an ending so that the for loop repeats the speci-
fied cycles. The program can be written as in the following example.

Figure 13. Iteration using the for command, for Loop 1

In the figure, the data scientists assign a range of data determining the variable n as 5, and then uses
a for loop to render the term “Digital Transformation” for 5 times. In such this loop, the initialization
does not need to be set, and the routine will repeat according to the end of the configuration cycle.

218

Data Mining for Junior Data Scientists

for Loop 2: Iteration to show values inside variables

for i in variable:
command

Iterating with the for command in for Loop 2 method, data scientists can iterate over the amount of
data within variables, and can write the program as follows.

Figure 14. Iterating using the for command, for Loop 2 method

In the figure, the data scientists pack 1 – 5 data into variable n, then using the for command to display
the value inside variable n for 5 times, according to the number within variable n. The data scientists
can also use the loop in text rendering where the command set will operate according to the loop cycle
equal to the number within the variable.

b. While Loop

When data scientists intend to perform the iteration while the conditional result is True, the while
command can be used for iteration with the following syntax.

Initialization of variables
while variable default condition:
command
Increasing or decreasing the value of the initialized variables

219
 
Data Mining for Junior Data Scientists

From the syntax, the data scientists initialize the variables used for iteration control. Then the while
command is used to check the condition. If the result of the condition is True, the given command is
executed. Then the value within the variable is increased or decreased in order to check the condition
again. If the result of the conditional check appears to be False, it exits the loop. The program can be
written as in the example below.

Figure 15. Using the while statement

As in the example, data scientists initialize variable i. The while command then checks if the value
inside variable i is less than or equal to 10. If the result is True, it displays the value inside variable i
through the print command. Then the variable i value is increased in order to create loops until the value
inside variable i is not greater than or equal to 10. If the result of the condition is False, it will exit the
iteration.

c. Exiting the loops with the Break command

When data scientists use the while command for iterative operations. If the data scientists intend to
exit the loops, the break command can be used as in the following syntax.

while True:
command
adding the iterative value
if condition:
break

220

Data Mining for Junior Data Scientists

From the syntax, data scientists can run a digital device to iterate while the value obtained through
the while command is true, and incrementally increment the value to perform iteration. To exist the
loop command is when the while command is checked and the value is False. If data scientists intend to
exit the loop by adding conditions, they can create conditions with if command and then use the break
command to exit the loop if the value checked through the if command is True. A program can be writ-
ten as follows.

Figure 16. Exiting the loops with the break command

In this example, the data scientists initialize the variable i to 1 by repeating the value inside the vari-
able i while the value inside the variable i is less than or equal to 10. The value is continually increased
until the value inside i is obtained as False. However, data scientists can exit the loop with the break
command when the value inside the variable i equals 5.

d. Excluding processing within the loop with the Continue command

When data scientists use the for loop command, some loops can be skipped. If the aforementioned
loops meet the defined conditioned, to skip, they can use the continue command as in the following syntax.

if condition:
continue
command

221
 
Data Mining for Junior Data Scientists

From the syntax, the data scientists can skip the execution of the command set if the value received
from the condition is True. And they can skip the execution of the command with the continue com-
mand, and execute the command when the condition is false. The program can be written as follows.

Figure 17. Skipping the loops with the continue command

As in the example, the data scientists use the for loop to represent numbers from 0 to 9. It iterates to
represent the value in variable i only if the value obtained in the for loop is True, which is in the speci-
fied data range. And the continue command is used to skip the execution of the commands when the
value inside the variable i is 5. Therefore, when executing the command set, displaying numbers 0 to 9
except 5. It corresponds to the exception of executing the command set using of the continue command.

BASIC OF DATA STRUCTURE

When data scientists intend to store multiple data within a single variable; importing a data set from a
computer into Python command set, each variable needs to put a large amount of data into a single vari-
able. Scientists need to manage and access data within different types of data structures.

List

Data scientists can pack multiple amounts of data into List variable. The syntax for creating List is as
follows.

222

Data Mining for Junior Data Scientists

variable = [data, data, …, data]

From the syntax, a single variable acts as List by storing multiple data in square brackets. Data sci-
entists can create List as follows.

Figure 18. Creating list

As in the example, data scientists can create the variable i to store all five numeric data; consisting
of 1, 3, 5, 7, and 9 respectively, and then display them via the print command.

Managing Data Within List

Data scientists can use commands to access and manipulate data within List, either by adding data to a
list, replacing existing data or deleting data with the following syntax.

Syntax 1: Adding data to list, Method 1

variable += [‘data’, ‘data’]

According to Syntax 1, data scientists can use list variable to add new data into the list using symbol
+=. The newly added data will come after the previously added data.

Syntax 2: Adding data to list, Method 2

variable.append(‘data’)

223
 
Data Mining for Junior Data Scientists

According to Syntax 2, data scientists can add data to the end of the current data by using the .append
command. Data scientists can choose to use either syntax 1 or 2 in their programming.

Syntax 3: Replacing the existing data

variable[data location] = “data”

According to Syntax 3, when data scientists intend to replace existing data within list, they can refer
to the data location within the list and insert the new data into it.

Syntax 4: Deleting data within the list

variable[data range] = [ ]

According to Syntax 4, the data scientists can specify the location of the data to be deleted, and fill
in a blank value to replace that location. Then, the appended data of the deleted data will be replaced
the empty position.
Data scientists can program all the four syntaxes as follows:

Figure 19. Managing data within list

In the example, data scientists create variables that act as list called staffs, containing Jirapon data
in position 0, Sarawut data in position 1, Chayakrit data in position 2, and Kannapa data in position 3.
Data are then displayed through the print command.
Kom data is then added to position 4 and Noppol data to position 5 with +=, and the data is displayed
with the print command. The Kom and Noppol data are appended to the preceding data within List.

224

Data Mining for Junior Data Scientists

In the next operation, data scientists add Jiroj data to position 6 with the .append command, and
then display it via the print command. When the data scientists intend to replace the data in position 5,
formerly Noppol, with the data Jibby, so they can refer the position of the data and put the new data to
replace it and display it through the print command.
When the data scientists determine that the location of data to be deleted is positions 1 and 2. Next,
the data is then replaced. As shown by the print command, the deletion of an item from a list is performed
to the position before the deletion’s final digit.

Dictionary Key and Values

When data scientists want to provide a variable that packs multiple types of data into one variable, they
can write a program with the following syntax.

Syntax 1: Creating a dictionary

variable = {key: value}, …, {key: value}

According to Syntax 1, data scientists can create a dictionary consisting of a set of data that acts as
keys and values, both of which can be either numeric or alphabetic.

Syntax 2: Adding data into the dictionary

variable[key] = value

From syntax 2, data scientists can add key and value data to the dictionary by appending the existing
data.

Syntax 3: Removing data from the dictionary

del variable[key]

From syntax 3, data scientists can delete data within a dictionary using the del command, and refer
the keys needed to delete. After execution, the referred keys and values are deleted from the dictionary.

Syntax 4: Displaying data of all keys and values within a dictionary

variable.keys()
variable.values()

From syntax 4, data scientists can display all data in a dictionary. It displays all keys and all values
via the print command.
With these four syntaxes, data scientists can write programs to manipulate dictionaries as in the example.
As in the example, data scientists create a dictionary by filling the data into the students variable
consisting of 3 sets of letter keys and numeric values, and the result is displayed. Then, a new set of
data is added into the variables before displaying them again. The newly added data is appended to the
existing data.

225
 
Data Mining for Junior Data Scientists

Figure 20. Manipulating the dictionary

Data scientists can then delete the data from the dictionary by referring the keys using the del com-
mand. As seen in the example, the data scientists referred the Jirapa key, so after re-rendering the data,
the Jirapa key and the value 123 are then removed from the dictionary.
The data scientists can render all key data within a dictionary using .keys() and returns only the values
of all data within a dictionary using .values().

Searching for Data With Keywords

Where scientists store large amounts of data in variables that function as list or a dictionary, they can
search for the data using the following syntax.

Syntax 1: Searching for data from the list

“keyword” in variable

According to syntax 1, data scientists can search for data within a list and display it via the print
command.

Syntax 2: Searching for data from the dictionary

‘keyword’ in variable

Using syntax 2, the data scientists can search for data within a dictionary and display it through the
print command.

226

Data Mining for Junior Data Scientists

With both syntaxes, data scientists can program the search of data in the list and the dictionary as
follows.

Figure 21. Searching data with keywords

As in the example, data scientists create the list containing Sunkpho, Ramjan, and Campiranon data
respectively. Then, it searches for Ramjan within the staffs_list variable and displays it through the print
command, and the result is True, which means the data is contained within the list. Then a dictionary
is created. It contains data including Sunkpho:1, Ramjan:2, and Campiranon:3. Then Sunkpho data is
searched within the staffs_dict variable, which acts as a dictionary, and the result is True, which means
the data is actually contained in the dictionary.

IMPORTING DATA INTO A COMMAND SET (FILE I/O)

When data scientists intend to put data into a set of instructions for processing with various techniques,
they can use commands within Python to read the data by inputting data via the keyboard and creating
.txt format files.

Importing Data Into the Command Set via a Keyboard

In the case that data scientists intend to input data into the command set via a keyboard, they can write
a program with the following syntax.

227
 
Data Mining for Junior Data Scientists

variable = input(“alt text”)

From the syntax, data scientists can write a program to input data into the command set via the
keyboard as in the example.

Figure 22. Importing data into the command set via keyboard

In this example, data scientists input data into variables via a keyboard using the input command, where
number 20 is inserted into variable aa, and the value is displayed inside variable aa on the keyboard.

Creating and Manipulating Text Files

Data scientists can create a .txt, Text File, which supports writing data into that file. It can also bring
the written data into the file displayed on the screen. The syntax for writing and reading data within the
file is as follows.

Syntax 1: Writing data into the text file

file variable = open(“file name .txt”, “w+”)
file variable .write(“data”)
file variable. close()

Based on syntax 1, data scientists can create .txt file by naming the file and using w+ mode for writ-
ing data into text file with the open function. The file is then put into the file-holding variables. Data
scientists can use the .write() method to write data and the .close() method to close the file.

228

Data Mining for Junior Data Scientists

Syntax 2: Reading data from the text file

file variable = open(“file name .txt”, “r+”)
variable = file variable.read(character count)
print (variable)
File variable.close()

According to syntax 2, data scientists can open a file with open function, using r+ mode to read the
data from the Text File. The data are then imported from the Text File into a variable using the. read
method, and the data are displayed through Print command, The file can be closed with the close() method.
Data scientists can write a program to create a file, write data and read text file as shown in the
example below.

Figure 23. Creating a File, writing and reading data from text file

In the figure, data scientists can create a file named aa using the open function. And w+ mode is
used to write data into the Text File via the .write method. As in the example, data scientists can write
the Digital Transformation and Innovation data on the Text File, and then close the file with the .close()
detach method.
After that, data scientists reopen the aa file, use r+ mode to read the data in the text file, load the
file into the bb variable, and then execute the .read command, which reads all 40 characters. Later the
data are filled into the str variable, which sdisplay the data through the print statement. Then the file is
closed again with the .close() method.

229
 
Data Mining for Junior Data Scientists

EXCEPTION HANDLING

In case an error occurs during processing, it causes the program to stop working. When the command
set does not find the files or data required to process, data scientists can use Exception Handling to skip
the error, so that the command set can continue processing. The syntax is as follows.

Try:
command
except, type of exception:
command
else:
command

According to the syntax, data scientists put all commands to be processed inside try: After that, if
during processing occurs any error, process it inside except. And if there is no error, process it inside else:
The except processing occurs only when there is an error as specified; an IOError which occurs
the command set is unable to open the file, or SyntaxError which is caused by a syntax error whiling
writing a program. Within a command set, the except command can be written in a number of cases;
when checking for both IOError and SyntaxError after processing with try: Data scientists can write a
command set covered by Exception Handling, as in the example.

Figure 24. Using exception handling in case the command fails to open the text file

230

Data Mining for Junior Data Scientists

As in the example, data scientists write a command set with try: to open and read a text file called
dd.txt, which hasn’t been created before. Therefore, the IOError exception condition is met, so the com-
mand executes the print statement “I could not find a text file” and terminates it. In another case, if the
statement under try: is valid, it will be displayed as follows.

Figure 25. Using exception handling in case the command set can open files

As shown in the figure, the command can open Text File named aa.txt, which is a pre-written file.
The command is then executed after else: which is to display the statement “Everything Correct” and
close the file with the. close () method.

USING GOOGLE COLAB

Data scientists can choose from a wide variety of software editors either by installing on the computer
or in processing through the web such as Google Colab. Google Colab also has an Interpreter set ready
to translate any Python language with no additional setup required. And all files are stored on Google
Colab’s memory without the need to store them on a computer.
Data scientists can access Google Colab by registering for gmail such as [email protected], then
login to use Google services, then enter Google Colab through the link is https://colab.research.google.
com, and enter to open the file for programming as seen in the picture below.

231
 
Data Mining for Junior Data Scientists

Figure 26. Creating a programming file

From the figure, data scientists select NEW NOTEBOOK to create a file for programming. And if
intending to continue working from the created file, they can select the file name that appears or search
from Google Drive and start working, as shown in the picture below.

Figure 27. Getting started with Google Colab

232

Data Mining for Junior Data Scientists

Once data scientists enter a file created by Google Colab, they can write a command set in Python.
Then the command set is tested by pressing the Run Cell key or using the Ctrl + Enter key to execute
the command with the Interpreter in Google Colab. The result is presented below the command set.
To write additional commands without creating new files, data scientists can press the + Code button.

Figure 28. Writing additional commands

As in the figure above, data scientists can write multiple commands in a single file without affecting
previously programmed commands. Another outstanding use of Google Colab is the display of possible
commands. For example, if the scientists type Ex into Google Colab, it will show possible commands
for data scientists to choose, which is to reduce programming errors.
When data scientists intend to create a new file or access the file stored on Google Drive, they can
go to the file menu and select New Notebook to create a new file and select Open Notebook to access
the old file.
Once logging into Google Colab, data scientists can share the created files with other data scientists,
and those who have shared the files can also edit them along with those who own the original files. The
results of processing with Google Colab are very nice for visualization, making Google Colab a popular
software editor in data science.

233
 
Data Mining for Junior Data Scientists

Figure 29. Showing possible commands by Gogole Colab

Figure 30. Creating new files and accessing the old files

234

Data Mining for Junior Data Scientists

CONCLUSION

Data scientists can use Python’s as an open source to process big data while reducing the cost of digital
resource investments. Python also provides libraries, providing a set of commands for data processing.
As a result, Python has become a powerful tool in data science and a computer language that scientists
should study as a foundation for future studies. This book introduces the basic commands that data sci-
entists need to study before writing a program to put data into a set of commands, and displaying data
in various formats.
Google Colab is a software editor that can be used to write commands in Python. It also provides
libraries that store command sets for data mining processing. Data scientists can do experiments, write
command sets and test them via the web, without installing software on a computer.

EXERCISE QUESTIONS

1. Write a program in Python to get 2 numbers from the keyboard, then add, subtract, multiply, divide
and display the result through the screen.
2. Write a program in Python to get the pork weight from the keyboard, then multiply the weight by
$2 to calculate the pork price that the customer has to pay the pork shop.
3. Write a program for a hamburger restaurant to serve customers.
◦◦ If the customer wants a beef hamburger, press number 1.
▪▪ Have the customer enter the number of hamburgers on the keyboard. Then multiply the
number of hamburgers by 4 dollars and show the price that the customer has to pay on
the screen.
◦◦ If the customer wants a pork hamburger, press the number 2.
▪▪ Have the customer enter the number of hamburgers on the keyboard. Then multiply the
number of hamburgers by 3 dollars and show the price that the customer has to pay on
the screen.
◦◦ If the customer presses another number, it will alert “Choose between numbers 1 and 2”.
4. Write a program to get scores from the keyboard.
◦◦ If the score is between 80 - 100, show the letter A on the screen.
◦◦ If the score is between 70 - 79, show the letter B on the screen.
◦◦ If the score is between 60 - 69, show the letter C on the screen.
◦◦ If the score is between 50 - 59, show the letter D on the screen.
◦◦ If the score is between 0 - 49, show the letter F on the screen.
◦◦ If the score is more than 100 or less than 0, display a message “Please enter your score from
0 – 100”
◦◦ If the user wants to redo, press Y, and to exit, press Q using While Loop

REFERENCES

Bader, D. (2018). Python tricks: The book. Dan Bader.

235
 
Data Mining for Junior Data Scientists

Bowles, M. (2015). Machine learning in Python: essential techniques for predictive analysis. Wiley.
doi:10.1002/9781119183600
Hill, C. (2015). Learning scientific programming with Python. Cambridge University Press. doi:10.1017/
CBO9781139871754
Joshi, P. (2017). Artificial intelligence with Python: a comprehensive guide to building intelligent apps
for Python beginners and developers. Packt Publishing Ltd.
Lubanovic, B. (2019). Introducing Python: modern computing in simple packages (2nd ed.). O’Reilly
Media, Inc.
Mueller, J. (2018). Beginning programming with Python (2nd ed.). John Wiley & Sons.
Raschka, S., & Mirajalili, V. (2018). Python machine learning: machine learning and deep learning
with Python, scikit-learn, and TensorFlow (2nd ed.). Packt Publishing.
Reges, S., Obourn, A., & Stepp, M. (2019). Building Python programs (1st ed.). Pearson.
Sneeringer, L. (2016). Professional Python. WROX/John Wiley & Sons.
Stewart, J. (2017). Python for scientists (2nd ed.). Cambridge University Press. doi:10.1017/9781108120241
Sunkpho, S., & Ramjan, S. (2020). Introduction to Python Programming for Data Science. Thammasat
Press.
Tananatong, T., & Ramjan, S. (2022). Data Visualization Design and Development for Data Science.
Thammasat Press.
Zelle, J. M. (2010). Python programming: an introduction to computer science (2nd ed.). Beedle &
Associates.

236
 237

Chapter 12
Data Mining for Junior
Data Scientists:
Data Analytics With Python

ABSTRACT
It is crucial for junior data scientists to learn computer programming as data science software packages
may not always cater to the requirements of data analysis. Python provides a vast library of algorithms
for data analysis, including NumPy, Pandas, Matplotlib, Seaborn, and Scikit-learn. NumPy and Pandas
aid in organizing datasets as part of the pre-processing stage, while Matplotlib and Seaborn offer a
range of data visualization commands. These visualization tools are instrumental in data exploration
processes, such as creating histograms and scatter plots, and displaying data mining results like cluster
analysis outcomes. Scikit-learn is a popular library in the data science industry that offers various data
mining commands for regression, decision constructs, and cluster analysis, covering both supervised
and unsupervised learning. Therefore, junior data scientists must learn Python programming for data
science applications, especially when using software packages that require editing the model using
Python commands.

INTRODUCTION

Nowadays, there are tools used for analyzing data during the workflow; the data extraction tools, data
survey tools, data preparation tools, Data Analysis Tools with Data Mining Techniques and Data Visual-
ization tools. Each step has a ready-to-use software in both instant software and language programming
software. Python is a tool that supports the entire workflow as it can be used for data analysis purposes;
manipulating datasets, importing routines, developing data visualization, and data analysis with data
mining techniques using libraries (Massaron & Mueller, 2015: Mueller & Massaron, 2019).

DOI: 10.4018/978-1-6684-4730-7.ch012

Copyright © 2023, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
 
Data Mining for Junior Data Scientists

NumPy Library

The NumPy library provides a set of commands that data scientists can execute to manipulate imported
datasets and then process them within Python’s command set. The library focuses on compiling data
into the format ready to process with data mining techniques:

Array

Python supports storing data in arrays, which are in rows and columns. Array variables can store data
of one to more dimensions. The structure is as follows:

Table 1. Dimension array

Column 0
Row 0 0,0
Row 1 1,0
Row 2 2,0
Row. .
Row. .
Row n n,0

From the table, data scientists can store data as a 1-dimensional array with records starting from 1
row and 1 column. Inside array variables, the location of the data is specified at row 0 and columns 0.
For example, the number 2 might be contained in row 0 and column 0. When data scientists store more
than 1 column, a 2-dimensional array variable can be created with the following structure:

Table 2. Dimension array

Column 0 Column 1 Column 2 Column . Column . Column n

Row 0 0,0 0,1 0,2 0,. 0,. 0,n
Row 1 1,0 1,1 1,2 1,. 1,. 1,n
Row 2 2,0 2,1 2,2 2,. 2,. 2,n
Row . .,0 .,1 .,2 .,. .,. .,n
Row . .,0 .,1 .,2 .,. .,. .,n
Row n n,0 n,1 n,2 n,. n,. n,n

From the table, data scientists can store data in multiple rows and columns simultaneously. For ex-
ample, the numeric data 25 may be positioned in Row 5, Column 3. By storing data in this manner, the
data scientists can design data structures with good memory management.

238

Data Mining for Junior Data Scientists

Therefore, when manipulating with datasets that require array variables to store data, data scientists
can run commands from the NumPy library with the following syntax:

import numpy as the variable that executes command sets

Array variable = the variable that executes the .array command([data,data..data])

From the syntax, data scientists can invoke .array commands from the NumPy library to create array
variables, which can be programmed as in the example.

Figure 1. Using commands to create array variables

In the figure, the data scientist invokes an .array command from NumPy library where variable mm
executes the command to generate array data of 5, 6, and 7, respectively. Then the data are packed into
variable a and displayed via print command
When data scientists intend to store data as a 2-dimension array, they can also use the .array com-
mand from the NumPy library. The syntax is as follows:

import numpy as the variable that executes command sets

Array variable = the variable that executes the .array command([(data,data..data), (data,data..data)])

From the syntax, data scientists can add the second data. In Python, the data inside the parentheses
is counted as 1 row.

Figure 2. Using commands to create a two-dimensional array

239
 
Data Mining for Junior Data Scientists

In the figure, data scientists use variable mm to invoke the .array command from the NumPy library
and generate two datasets; 5, 6, 7 as in set 1, and 8, 9, 10 as in set 2. Then, the data is put into variable
a and displayed through the print command.

Creating Array Data With .arange

Within the NumPy library, there is also a set of .arange commands that data scientists can use to create
a series of numbers, the array variables, for other usages. The syntax is as follows:

import numpy as the variable that executes command sets

print(variable that executes the .arange command(number of numbers))

Based on the syntax, the data scientists get the variable to invoke .arange command from the NumPy
library and then specify a number to display via the print command. The set of numbers starts from 0 to
the previous of the specified number. The data scientists can write a program as in the example.

Figure 3. Creating an array dataset with arange

In the figure, the data scientists get an np variable to invoke .arange command from NumPy library,
then create a dataset of 5 arrays starting from 0 – 4 and display them via the print command.

The .reshape Command

After data scientists have created an array dataset with .arange command, they can arrange the data into
the specified row and column sizes by using .reshape command from NumPy library, as in the follow-
ing syntax.

import numpy as the variable that executes command sets

print(variable to execute .arange(numbers).reshape(numbers of rows, numbers of columns)

After creating a dataset with .arange command, data scientists can adjust the data into the specified
row and column format using .reshape command.

240

Data Mining for Junior Data Scientists

Figure 4. Compiling data with reshape

By creating a dataset containing 10 numbers, from 0 to 9, with .arange, the scientists compile the data
with .reshape command to have 5 rows and 2 columns, pack the data into variable a, and then display
it via print command. In order to compile data with .reshape command, data scientists must take into
account the data size.

The .linspace Command

Not only the .arange statement can be used to create an array of datasets, the .linspace command from
NumPy library can also be used to create periodic datasets.

import numpy as the variable that executes a set of instructions

the variables that execute .linspace (initial number, end number, data frequency)

From the syntax, the data scientists invoke .linspace command, defining the initial number and the
end number of the created datasets, and defining the frequency of the data amount to be generated in
the data range. The program is written as follows.

Figure 5. Creating a dataset with linspace command

In the figure, data scientists use np variable to invoke .linsapce command from NumPy library and
generate a data range starting from 7 to 15 with the frequency of 3 times. So, the created data are 7, 11
and 15, respectively.

241
 
Data Mining for Junior Data Scientists

Changing Data Types in Array Variables

When data scientists prepare the array data, they can use dtype command to change data types within
array variables. If the data type is an integer to be changed to a real number, it can be specified as float.
If the data type is a real number to be changed to an integer number, it can be specified as an int with
the following syntax:

import numpy as the variable that executes a set of commands

array variables = the variables that execute .array([[(data,data..data), (data,data..data)]], dtype=
‘data type’)

From the syntax, after the scientists create an array-like dataset, they can change the data type with
dtype by writing a program as in the example below.

Figure 6. Changing the data type

Based on the figure, data scientists create an array dataset where the internal data is all real numbers,
then use dtype command to change a real number into an integer. When checking using dtype command
via print command, the result is int64, meaning the data inside the variable have been converted into an
integer, and the entire results are re-rendered through print statement.

Mathematical Operations Within Array

When data scientists intend to transform data with mathematical operation methods, they can take array
variable containing a dataset to perform the operations as in the following syntax.

import numpy as the variable that executes command sets

Array variables = the variable that executes .array([data])
Array variables = the variable that executes .array([data])
variable = array variable mathematical operators array variables

242

Data Mining for Junior Data Scientists

From the syntax, the scientists can take the created array variables to perform the mathematical
operation such as addition, subtraction, multiplication, and division by writing the following program.

Figure 7. Mathematical operation of array variables

After the data scientists create the 1st array variable containing the datasets 7 and 4, and create the 2nd
array variable containing the datasets 0 and 1, the data in the array variables are mathematical operated,
subtracting, through the variables containing the two datasets. The result is then added into variable c.
Number 7 is to subtract with 0 to get the result as 7, and number 4 is to subtract with 1 to get the result
as 3, respectively.

Creating Array Variables With .random Command

Data scientists can also create datasets with .random from NumPy library with a range of 0 – 1 as in the
syntax below.

import numpy as the variable that executes the command

variable = the variable that executes the .random.random command([number of rows, number of
columns])

From the syntax, the scientists can create datasets with values ​​between 0 and 1 within array variable
using .random command, and then specify the number of rows and columns. Data scientists can experi-
ment with programming by the following example.

Figure 8. Creating a dataset with random command

243
 
Data Mining for Junior Data Scientists

In the figure, the data scientists use np variable, invoke .random command from NumPy library, create
a dataset ranging from 0 to 1 with 3 rows and 4 columns, and then output the data via print command.

Summarizing Array Data

After data scientists have created array datasets, they can summarize that data with commands such as
.sum to combine numeric data within array variable; .min to find the smallest number within thearray
variable, and .max to find the largest number in the array variable, as seen in the following syntax.

import numpy as the variable that executes a set of commands

Array variable = the variable that executes .array([data,data..data])
print(array variable.sum())
print(array variable.min())
print(array variable.max())

From the syntax, data scientists can sum up and find the largest and the smallest values from the
array variable. The program can be written according to the example.

Figure 9. Summarizing data in array variables

In the figure, data scientists create array variable consisting of 1, 3, 5, 7, and 9, respectively, then
use .sum command to sum the data, and get the result as 25. The .min command is then used to find the
smallest value which is 1, and the .max command is used to find the largest value which is 9.

Indexing and Slicing

When data scientists intend to access data by indexing, they can refer to the location to store the data,
and the can sort the data by slicing it according to the following syntax:

import numpy as the variable that executes a set of commands

Array variable = the variable that executes .array([data,data..data])
Array variable[location of data]

244

Data Mining for Junior Data Scientists

Array variable[start position:end position:display frequency)

From the syntax, the scientists can refer the location in order to access the data within that location
and manipulate them by displaying in intervals. The program can be written as in the example.

Figure 10. Accessing data locations and intermittent displays

Based on the figure, the scientists create array variable. The data consist of 5, 4, 3, 2, and 1 respec-
tively, and specify the location in position 3, which is the number 2. The result is displayed through
print command.
Then, the data is accessed in positions 1-7, and the output is displayed in every 2 positions, resulting
via print command as 4, 2 and 5 respectively.

10. Vectorization

When data scientists perform Attribute Transformation in the preparation process, they can use .vec-
torize command from NumPy library. Vectorization is to use small data called Broadcast to perform
calculations on the dataset in order to change the data. The syntax is as follows.

import numpy as the variable that executes the command

variable = the variable that executes .vectorize(computing function)

From the syntax, the data scientists invoke .vectorize command to calculate a small dataset. The
program can be written as in the following example.
In the figure, the data scientists get np variable to execute vectorize command from NumPy library.
Inside the command, the abc function is executed, packing all the data into the mm variable, then passing
the values 4, 3, 2, and 1 to the d parameter, and passing the value 2 to the e parameter. If the value inside
variable d is greater than the value in variable e, then subtract the returned values. If it is not greater,
then add them. And the result of the calculation is 2, 1, 4 and 3, respectively. In this way, the vectorize
command can be used to process the data.

245
 
Data Mining for Junior Data Scientists

Figure 11. Vectorization

PANDAS LIBRARY

When data scientists manipulate and process data with mining techniques, Pandas is a critical library
used for storing command sets.

Series and Data Frames

Whenever data scientists import data or create datasets within Python, the datasets will be then called a
Data Frame, and the columns within the dataset are so called Series. Data in each raw are called Instances,
and individual data is called Data Object. Data scientists can create a Series with the following syntax:

Import pandas as series variables

Series variables.Series([data,data, ….])

From the syntax, the scientists invoke .Series command to generate a single column of data by writ-
ing the following syntax.

Figure 12. Creating a series

246

Data Mining for Junior Data Scientists

In the figure, data scientists use the pd variable to invoke a .Series command from the pandas library
to generate the data within the Series. Then, when the scientists intend to create multiple Series and
combine them into a Data Frame, they can writing the program as in the following syntax.

Import pandas as the variable that invokes commands

variables that invoke .DataFrame({ column name: series variable})

From the syntax, data scientists can use .DataFrame command from the pandas library by naming
the column and the generated series variable. The program can be written as follows.

Figure 13. Creating a data frame

In the figure, data scientists have created a series of two columns containing the data on the three
courses and the number of students in each academic year. Then, the .DataFrame command is used to
name the Series and connect it with the Series, which is created by referring to variables that store the
Series data and combining all the Series into a Data Fame.

Importing Data Into the Command Set

The Pandas Library supports data import by importing computer dataset into Python and by importing
an Internet dataset into Python command set. To import data from a computer into a command set can
be programmed according to the following syntax:

import pandas as the variable that invokes commands

Data Frame variables = Variables that invoke csv(‘file source address’)

From the syntax, data scientists can take .csv files by accessing the file address and then importing
the data into the data frame variables. They can be programmed with Google Colab as follows:

247
 
Data Mining for Junior Data Scientists

Figure 14. Importing data from computer into command set

In the figure, the scientists select Upload to Session Storage, and then Google Colab enters the file
address. The scientists can then choose the file intended to import into Google Colab. In the example,
the file named chol.txt is chosen.
From the figure, data scientists can right click on chol.txt file and bring the file address to write the
program as shown in the example.
In the figure, data scientists use .read_csv command to import the dataset from chol.txt file into Google
Colab, then pack all datasets into cc variable and display the data via the print command.
In case the scientists intend to import any dataset available on the Internet, they can import the file
address into Google Colab without uploading it. The program can be written according to the following
syntax.

import pandas as the variable that invokes commands

Data Frame variable = Variable that invoke csv(‘Internet file source address’)

From the syntax, data scientists can refer to the source address of a dataset on the Internet to process
the data by writing a program according to the example.
In the figure, data scientists copy the address of the dataset using read_csv command from the Pan-
das library.

248

Data Mining for Junior Data Scientists

Figure 15. Entering the file address

249
 
Data Mining for Junior Data Scientists

Figure 16. Importing data from a computer into Google Colab

Figure 17. Importing data on the Internet into datasets

Indexing, Selecting, and Filtering

After data scientists import a dataset into Python’s programming language, the scientists can begin ma-
nipulating the data by referring to each Series within the Data Frame according to the syntax.

variable = the variable that executes command [“named Series”]

From the syntax, data scientists can refer to the name Series within a data frame, then place all In-
stances from that Series into variables and render them as in the example.

250

Data Mining for Junior Data Scientists

Figure 18. Data accessibility (indexing)

The figure shows how the scientists can be indexing or accessing the series named median_income in
order to display the data from all Data Frames. In other cases, the scientists can access multiple Series
from the same Data Frame in the meantime.

Figure 19. Selecting data

According to the figure, data scientists can add the Series intended to display using an extra layer
of square brackets so that they can cover multiple Series in the meantime. As in the example, the data
scientists access the Series including median_income, toal_bedrooms, and population.

251
 
Data Mining for Junior Data Scientists

When displaying all data, the scientists can also perform filtering to select only the data involved in
processing. The program can be written as presented in the following example:

Figure 20. Filtering

From the figure, the scientists are able to filter data to select only instances with the Series named
population greater than 1,429. From the sample, the number of instances has dropped from a total of
3,000 to 1,103.

Data Processing in the Data Frame

Data Processing Within the Data Frame

When data scientists intend operate the data within the Series and the variables that act as broadcast or
small data, they can access that series variable and create Broadcast variables and perform mathematical
operations by writing the following program.
In the figure, the scientists access the Series named population, create a Broadcast variable with a
value of 1000, place it inside variable b, then take all Data Objects within the Series to subtract from
variable b, and present the result through print commands.

Sorting Data Within a Data Frame

In the data survey process, data scientists can sort all instances within a dataset with a particular Series
to explore trends and patterns in the data. The program can be written as in the example.
In the figure, data scientists can use .sort_values command to sort instances from a particular series
to explore data patterns and trends. Based on the example, all data within the dataset are sorted according
to Series named median_hosue_value, sorting from the smallest to the largest number.

252

Data Mining for Junior Data Scientists

Figure 21. Using Broadcast along with series variables

Figure 22. Sorting instances with series

Dealing With Missing Values

Before analyzing data with data mining techniques, data scientists need to survey and prepare data, by
considering whether the Data Objects are missing values. Data scientists can write a program to detect
data objects that are missing values, then perform manipulation of the missing values, such as eliminating
the Data Object or filling the incomplete Data Object with another value. The program can be written
according to the following syntax.

253
 
Data Mining for Junior Data Scientists

import numby
from numpy import nan as NA

Based on the syntax, data scientists run NaN command from the NumPy library, working with other
commands from the Panda library to check for data containing NA or the missing values, and then
replace those values with NaN as in the example.

Figure 23. Verifying missing values

In the figure, the data scientists run the NaN command from the NumPy library. When examining data
containing NA text or the missing value and replacing it with NaN as seen in the sample, the scientists
create a total of 3 series, each of which contains data in the form of NA, then create a Data Frame from
that series and display it. All data in NA format are replaced with NaN.
Programming in such a way allows data scientists to determine whether there is missing value within
the dataset. To survey and prepare data, data scientists can consider dealing with those missing values,
such as eliminating incomplete data from the dataset or complementing such data with other values. It
can be programmed according to the syntax.

variable = dataset variable.dropna()

variable = variable.reset_index(drop=True)

254

Data Mining for Junior Data Scientists

From the syntax, data scientists use .dropna() command to remove NaN-formatted data. Then, the
.reset_index(drop=True) command is used to rearrange the purged dataset, which can be written as in
the example.

Figure 24. Eliminating incomplete data

In the figure, data scientists extract the NaN-formatted data from the Data Frame contained in the data
variable, and then use the .dropna() command to remove the NaN- data from the data frame. The data
frame is then loaded into the dataclean variable. At the next step, the scientists reformat the data frame
using .rest_index(drop=True) and leave only the complete data command. It is then rendered for a result.
In this case, the data scientists determine that the incomplete data should be filled, such as replac-
ing with the mean or substituting with nominal values. The syntax for filling incomplete data with the
mean is as follows.

variable = fillna(data.mean())variable.

From the syntax, data scientists use the .fillna(data.mean) command to fill incomplete data with the
mean, which can be programmed as in the example.
In the figure, the data scientists generate an average from the mean of each series, and then fill that
average into the missing values of each series by running the fillna.(data.mean) command and then
display all data.

255
 
Data Mining for Junior Data Scientists

Figure 25. Filling missing values with the mean

Dealing With Duplicate Data

In the process of surveying and preparing data, another thing that the scientists need to consider is how
to deal with duplicate data. Therefore, the first step to take is to verify whether the dataset, imported into
the Python command, contains duplicate data. It can be written with the following syntax.

variable.duplicated(keep=False)

From the syntax, data scientists can use the .duplicated(keep=False) command to check whether the
data object inside the data frame is duplicate or not. If the value is True, it is duplicate. And if the value
is False, it means that it is normal. The program can be written as follows.
In the figure, after data scientists create a data frame from a series, check for duplicate data using
the .duplicated(keep=False) command, display the result through the print statement, and execute the
dataset, they have found a total of 4 duplicates. The duplicate value is shown as True.
After the scientists examine the duplicate data, in the data preparation process the data scientists can
reduce this duplication while maintaining the integrity of the data frame structure. The program can be
written as the following syntax.

variable.drop_uplicates(inplace=True)

256

Data Mining for Junior Data Scientists

Figure 26. Verifying duplicate data

From the syntax, data scientists can use the .drop_duplicates(inplace=True) command to reduce
duplicate data. The program can be written as in the following example.

Figure 27. Reducing duplicate data

In the figure, data scientists reduce data duplication by using .drop_duplicates(inplace=True) com-
mand. After executing the command set, duplication is reduced, but the structure of the data frame is
still preserved. The aim is to reduce the duplication of the data that lack integrity to be analyzed.

Dealing With Outliers

Data scientists can investigate outliers from the datasets imported into Python command, such as
excessive-valued data or deficient-valued data. These data can cause the average value of the entire
feature to change, affecting the patterns and trends of the analyzed data, leading to the inaccuracy of
the data analysis.

257
 
Data Mining for Junior Data Scientists

Thus, data scientists can compute z-values, which reshape the entire data object’s feature to the form
of data under Normal Curve. If the data falls within the Standard deviation (SD) of over 3 or – 3, those
data objects are considered outliers. You can write a program as in the example below.

Figure 28. Detecting outliers

Data scientists import data into a command set and check the data object inside the data frame to
verify whether it is an outlier when the data is transformed to z-value and it is in the normal curve region
with a standard deviation greater than 3. In other words, the program does not use commands from the
Python library; rather it is programming based on the normal z-value algorithm. And then the result
value is checked whether it is greater than 3. And if the scientists intend to check whether the Data
Object is deficient-valued or not, this can be done by changing the number 3 to – 3 and re-establishing
the verification conditions.
Finally, the command set displays only instances with the z value greater than 3, so that the scientists
can determine whether the value is outliers. However, the scientists should consult with a knowledge
expert to consider the context in order to reassure.

258

Data Mining for Junior Data Scientists

Developing Data Visualization With Matplotlib and Seaborn Libraries

Data scientists can develop data visualizations in response to data survey purposes and data presentation
(Healy, 2018: Pajankar, 2021). By developing data visualizations, data scientists can explore patterns of
data both in terms of trends and distribution. Within Python, the Matplotlib and Seaborn libraries, whose
commands the scientists can use to develop useful visualizations for data exploration, are provided.

Line Chart

Data scientists can create line graphs made up of two data sets using the .plot command from the Mat-
plotlib library.

import matplotlib.pyplot as variable that executes a command set

Variable that executes the .plot([axis data x],[y-axis data])
Variables that run the .show() routine.

From the syntax, the scientists get the variable to run .plot command in the matplotlib.pyplot library
in order to create a line graph. Then the result is displayed with the .show() command. You can write a
program to create a graph as follows.

Figure 29. Creating a line graph

In the figure, data scientists use the variable cc to run the .plot command in the matplotlib.pyplot
library, then prepare the data for the x and y axes to create a line graph, and display it through the .show
() command. In case data scientists intend to develop a line graph by adding the visual guideline and
describing each part of the graph, they can write the program as in the syntax below.

259
 
Data Mining for Junior Data Scientists

import matplotlib.pyplot as variable that executes a command set

from matplotlip import style
style.use(“ggplot”)
Variable that executes .plot([axis data x],[y-axis data])
plt.title(“Graphic label”)
plt.xlabel(“x-axis alt label”)
plt.ylable(“y-axis alt label”)

Based on the syntax, the scientists import the style command from the matplotlib library and assign
the ggplot format in order to run the plt.title command to create the graph label, run plt.xlable com-
mand to create the x-axis label, and run plt.ylable command to create y-axis label as well as the visual
guideline. The program is as mentioned in the following example.

Figure 30. Creating a ggplot patterned line graph

From the figure, data scientists can use the ggplot pattern to create guide lines, and provide the
graph titles and labels for the x- and y-axis. From the graph, the scientists can also apply quantitative
data that is presented based on clustered data. For instance, it can be used to show the relationship of
the incidence of COVID-19 with the number of tourism, which is quantitative data presented in 2020
and 2021 as follows:

260

Data Mining for Junior Data Scientists

Figure 31. Presenting quantitative data by clustering

In the figure, data scientists create two data clusters, year 2020 and 2021, to compare the correla-
tion of COVID-19 with tourism rates. In each data set, a label can be specified to assign data to the plt.
legend() command, which is used to create data labels. The line graph for 2020 is orange and the line
graph for 2021 is blue.

261
 
Data Mining for Junior Data Scientists

Bar Chart

Data scientists can develop bar charts from quantitative data combined with data used for classification.
The program is as mentioned in the following syntax:

import numpy as nv
import matplotlib.pyplot as a variable that runs the command to create a graph
Series acting as the x-axis = [data]
Series acting as the y-axis = [data]
variable that control graphing = variable that executes the graphing command (len(Series acting as the
y-axis)
variables that run commands to create graphs .bar(variables that control graph creation, Series that act
as axes)
variables that run commands to create graphs .xticks(variables that control graphing, Series acting on
the x-axis)
variables that execute the command to create graphs .show()

From the syntax, data scientists can create a Series for the x and y axes, and then use the .arange(len())
command to count the number of data from Series acting as the x-axis. Data scientists can create quanti-
tative data using the .bar command and divide the data display by its type using the .xticks() command,
and then render it using the .show command. The program can be written by the following example.
In the figure, the data scientists run the .arange command from the NumPy library to count and
compile a bar chart with data-classifying variables. Then, the variable p runs a command to create a bar
graph from the matplotlib.pyplot library. Once data scientists have generated data for the x-axis which
is the types of animal and the y which is the quantity of each animal, they then create a bar based on the
data in the y-axis, display them by types of data in the x-axis via the .show() command.
When data scientists intend to add description to the bar chart, they can write a program as in the
example.
Data scientists can add more-detailed description to the graph. The graph can be named using .title
command, then the x-axis can be described with p.xlabel command, the y-axis can be described with
p.ylable command, and finally the bar description can be created with p.legend() command.
In cases that the scientists intend to display a comparison of the quantitatively-divided data and the
data used in two classifications; for example, the number of Zoo animals in 2020 and 2021, and animal
species, they can write programs to display the result as follows.
In the figure, data scientists create two data sets of: number of animals according to their types in
2020 and 2021. The year 2020 is represented in green as 2021 data is represented by purple. As the 2021
data is designated in purple and the color intensity is 0.5, the 2020 data is seen as being less volume and
green even though the bar showing the data for both 2020 and 2021 are displayed overlapped.

262

Data Mining for Junior Data Scientists

Figure 32. Creating a bar chart

Pie Chart

When data scientists intend to make a quantitative comparison of data in percentage by dividing the
display by data types. The program to create a Pie Chart can be written with the following syntax:

import matplotlib.pyplot as variables that execute the command

storage variable for categorization = [data used for categorization]
quantitative storage variables = [quantitative data]
variables that use pie command (quantitative storage variables,
labels = storage variables for categorization
autopct = “data display in percentage”)
p.title(“chart explaining Data”)
p.show()

263
 
Data Mining for Junior Data Scientists

Figure 33. Adding details to the bar chart

From the syntax, the scientists create a variable that run the .pie command to create a pie graph from
the matplotlib.pyplot library. It then generates the categorization data and the quantitative data to create
a pie graph by using the quantitative storage variables, then packing the categorization storage variables
into labels, and displaying them in percentage with autopct.
Data scientists can name a graph with the p.title command and display the graph with the p.show()
command.
From the figure, the scientists generate the data for categorization and putting into the song variable,
and then generate the quantitative data to pack into the variable r. The pie graph aims to show the pro-
portion of the total data in percentage. Therefore, when the quantitative data and the classification data
are used to create a pie chart, the data are converted into a proportional form and the result is displayed
in percentage with the autopct command.

264

Data Mining for Junior Data Scientists

Figure 34. Creating two data sets to compare in the bar chart

The sample pie graph shows the ratio of popular classical songs in the world. In case the scientists
intend to focus on a particular proportion on a pie chart, they can write a program as follows.
From the figure, data scientists can determine the position on the pie graph further away from the
other proportions so that they can focus on the desired display. To create a pie graph, the distance of the
data proportion can be specified with explode. When the scientists intend to represent data types with
legends to communicate graph details, they can write a program according to the following example.
In this figure, data scientists can use the classification datasets to describe a pie graph with the .leg-
end() command. In the example, a Legend is created showing the color that represents the proportion
which it is classified with the datasets.

265
 
Data Mining for Junior Data Scientists

Figure 35. Creating a pie chart

Histogram

Histograms are used to explore the distribution of data where the x-axis is created from the frequency of
the data in a range, and the axis y is created from the quantitative values of the data of the same feature.
Therefore, the histogram is generated from the 1 feature data.

import matplotlib.pyplot as variable that executes the command

data containing variable = [data]
variable that executes .hist(data containing variable)

From the syntax, data scientists can insert data into variables and use .hist command from the mat-
plotlib.pyplot library to create a histogram. The following example shows how to write the program.

266

Data Mining for Junior Data Scientists

Figure 36. The focus display of a particular aspect in the pie chart

From the figure, data scientists insert the data set named california_housing_train.csv into the
.read_csv command from the pandas library, and then access the feature named median_income. Later
it is inserted into variable x to create a histogram with .hist command. In the example, the data frequency
range is created from the feature named median_income as the x-axis, and then the y-axis is created from
the quantitative data in the same feature. The scientists can determine the distribution of data during
the survey process.

267
 
Data Mining for Junior Data Scientists

Figure 37. Adding a legend to the pie chart

When data scientists intend to display a histogram in conjunction with a line dispersion display, the
Seaborn library provides .distplot command to create a graph with such a pattern with the following syntax.

import seaborn as graphing variable

variable = data
graphing variable.displot(variable, kde=True)

In the figure, the scientists construct a histogram showing the distribution of the data together with the
lines by inserting the data set named California_housing_test.csv into the command, then add a feature
named median_income to the variable x, and create a histogram with .displot command to explore the
distribution of data in both bar and line formats.

268

Data Mining for Junior Data Scientists

Figure 38. Creating a histogram

Density Plot

Not only histograms can be used to explore data, the Density Plot can also be used to explore the data
in order to determine the distribution of the data. The y-axis, which is created by the Density Plot, using
the Probability of the feature, is considered a measure of how data is displayed. Data scientists can write
programs in the following syntax.

import seaborn as graphing variable

variable = data
graphing variable.kdeplot(variable)

From the syntax, data scientists can create Density Plot from the .kdeplot command from the seaborn
library.
From the figure, data scientists can create Density Plot for data exploration by having the mi variable
run the .kdeplot command from the seaborn library. As in the example, the california_housing_train.
csv dataset is imported into the command set, and the feature named median_income is inserted into the
x variable to create Density Plot. The x-axis is created from the data frequency, and the y-axis is from
Probability of the same feature.

269
 
Data Mining for Junior Data Scientists

Figure 39. Creating a density plot along with a histogram

Creating Box Plot

Data scientists can preliminarily explore five data: Min Value 25% Value, Mid Value, 75% Value, and
Maximum Value, from the explored feature. Such data can also be expressed in the form of Box Plot,
which displays quantitative data along with the classification data.

import seaborn as variable that executes a command set

storage variable = command executing variable(data).
data.boxplot(x= ‘x-axis Feature’, y= ‘y-axis Feature’, data = storage variable, order = [display of each
type]

From the syntax, data scientists can write a program to create Box Plot using the .boxplot command
stored in the Seaborn library, and then specify the feature used for creating the x-axis, which is the clas-
sification data used for identifying the feature to create the y-axis, which is quantitative data. The dataset
is then taken into the Box Plot and finally rendered.

270

Data Mining for Junior Data Scientists

Figure 40. Creating density plot

Figure 41. Creating box plot

271
 
Data Mining for Junior Data Scientists

In the figure, the scientists retrieve a sample dataset in Python called iris by running load_dataset
from the Seaborn library, loading it into the ee variable, and creating Box Plot. The x-axis displays the
feature named species as the names of Iris flower species, and the y-axis displays the feature named
sepal_length as the length of the Iris flower petal. The data from the ee variable are then entered into
the Box Plot, in the order of the Iris species; virginicia, setosa, and versicolor, respectively.
The Box Plot displays data for each Iris flower species, showing the outlier, the smallest value, the
25th percentile, the middle 75th percentile, and the largest value for the Iris width of each cultivar. It
allows data scientists to see the distribution of data before categorizing it with data mining techniques.

Scatter Plot

In the survey process, data scientists can use quantitative data from two or more features to create a
scatter plot to show the relationship between the two data. Scatter plots can also represent data that are
the outlier.

import seaborn as variable that executes the command

variable = variable that executes the command(data)
data.regplot(x= Feature that creates the x-axis, y =Feature that creates the y-axis)

From the syntax, data scientists can create a scatter Plot to determine the relationship of data using
the .regplot commands provided in the Seaborn library, graphing them from the quantitative data of
two or more features.
In the figure, the scientists create a scatter plot from two features, sepal_length and sepal_width,
contained within the iris sample dataset by running the .regplot command from the Seaborn library.
The Scatter Plot creates a data point from 2 features to show the relationship of data. Data scientists can
also determine the outlier from scatter plots.

FEATURE ENCODING

When data scientists insert a dataset into a command set to prepare the data for processing, they can use
Label Encoding and One Hot Encoding.

1. Label Encoding by adding feature names and monitoring for incomplete data.

In some cases, data scientists find that the datasets obtained from the data exploration process lack
the feature names to use for data analyzing programming, and those datasets may contain incomplete
data. The scientists can write a program to solve such a problem with the following syntax:

import pandas as the variable that executes commands

variable = [feature name data]
dataset storage variable = (header = None, names = variables, na_values= “?”)
the variable that holds dataset.head()

272

Data Mining for Junior Data Scientists

Figure 42. Creating scatter plot

From the syntax, data scientists use a variable to run the .read_csv command from the pandas library,
then assign the data within the variable as the feature in the dataset with names, and then assign the Data
Object within the dataset ? to be replaced with NaN. The program can be written as follows.

Figure 43. Adding a feature name and monitoring for incomplete data

273
 
Data Mining for Junior Data Scientists

In the figure, the data scientists use the pd variable to run the .read_csv command from the panda
library in order to insert the dataset imports-85.data into the command set. Then a dataset to be used to
name a Feature is created and put into the names so that the created dataset can be used as the feature
name. It is also determined as na_values =“?”, which means that whenever the data object which re-
sembles a question mark is encountered, the data object has to be replaced with NaN.
When displaying a dataset with .head() command, it is found that the data created to name the feature
is used for naming. However, before the data scientists name a feature, it is important to explore both the
number of feature and the nature of the data object in each feature for the great benefit of programming.

2. Label Encoding by converting letters to numbers

In many cases, data scientists receive datasets that are textual, used for classification such as gender,
province and region. However, data mining techniques use computation to analyze the data. Therefore,
data scientists have to turn those letters into numbers while retaining the classification properties. It can
be written as in the following syntax.

replacement data storage variable = {“Text Data Object”: Number}

dataset variable.replace(replacement data storage variable)

From the syntax, data scientists can identify a text data object that classifies the data and identifies
the numbers to replace the text so that those numbers remain classifiable. The program can be written
as follows:

Figure 44. Turning texts into numbers

274

Data Mining for Junior Data Scientists

In the figure, the scientists get the variable pd to run the .DataFrame command from the Pandas
library to create a dataset consisting of Cat, Lion, and Dog, respectively. Then, the data is packed into
the da variable and outputs it via the print command. The next step is to specify the number to replace a
text Data Object and put it into the convert variable. Then that variable is changed from text to number
with the .replace command.

3. One Hot Encoding

Changing text to numbers with Label Encoding cannot reflect quantitative data but still retain the
data properties for classification. One Hot Encoding is therefore used to count the number of text data
in the form of numbers, which can be used in quantitative analysis. It resembles Dummy data consisting
of the numbers 0 and 1. Data scientists can write a program according to the following syntax.

Import pandas as dataset variables

data storage variable = dataset variable.DataFrame({‘data category’:[data]})
dataset variable.get_dummies(data storage variable, prefix=[‘data category’])

From the syntax, data scientists can create Data Frame that stores data categories that cover all Data
Objects, and then quantify the text to be counted as the quantitative data. Then the Dummy is created
with numbers consisting of 0 which means no data, and 1 which means data. The program can be writ-
ten in the following example:

Figure 45. Creating dummy data set from text

275
 
Data Mining for Junior Data Scientists

In the figure, the scientists create a data frame, which contains the animal categories; Cat, Dog, Lion,
Lion, and Lion, respectively, and then present the result via the print command. When all data are created
from the animal categories in the form of a dummy dataset with the .get_dummies command, the result
is a table that counts the frequency of each format encounter. For instance, in the original data, Lion is
found 3 times. When it is changed into the form of dummy dataset, then the frequency of encountering
the word Lion is created as 3 of number 1. By creating the dummy data in this manner, Text data is
quantitatively valuable through counting the frequency of text occurrences.

4. Scikit-Learn Library

Python provides the Scikit Learn library, in which there is a command set for data mining. Data
scientists can insert the data set into the command set and then run a mining command from the
Scikit-Learn library to analyze it. Data analysis with machine learning (ML) can be divided into two
ways:

• Supervised-Learning: Is to divide the data set into 2 sets. The first set is used to teach the ma-
chine to recognize the nature of the data. The second set is used for testing machine accuracy by
Supervised-Learning, which is used in conjunction with various data mining techniques such as
regression, neural networks, and deep learning. When processing with data mining techniques, the
obtained results do not only forecast or classify data, but it can also show the accuracy of forecast-
ing and classification as well as compare the result with the actual data, resulting more closely to
the facts.
• Unsupervised-Learning: Is to bring data into data mining without teaching the machine to
learn. Data mining techniques are classified as Unsupervised-Learning, such as Clustering and
Association Rule Mining.

The Scikit-Learn library provides a command set for data mining in both forms; Supervised-Learning
and Unsupervised-Learning. Data scientists can experiment with the commands from

Scikit-Learn library as shown below.

Linear Regression Analysis is an analysis of the relationship of two features organized in a supervised
learning model where data scientists have to divide the data of the two features into two sets consisting
of the data for Train Set and Test Set (Sonu & Suyampulingam, 2021). The sample questions will put
forward the analysis of the relationship of height and weight by assigning the weight variable to the target
variable (Label). Data scientists can write a program according to the following syntax.

import matplotlib.pyplot as variable that runs the graph command

from sklearn.linear_model import LinearRegerssion
Feature variable = data
Target variable = data
Variable that invokes the graph command.scatter(variable Feature, target variable)
Variable that invokes the graph command.title(‘graph title’)
Variable that invokes the graph command.xlabel(‘x-axis label’)
Variable that invokes the graph command.ylabel(‘y-axis label’)

276

Data Mining for Junior Data Scientists

Data mining model variable = LinearRegerssion()

Data mining model variables.fit(X=Feature variable, y=target variable)
target variable = data mining model variable.predict([data to predict]])
print(target variable)
Variable that invokes the graph command.plot(Feature variable, model variable.predict(Feature variable))

From the syntax, data scientists can run regression analysis techniques with LinearRegerssion com-
mand from the Scikit-Learn library, then create feature to analyze data and target variables, and generate
a scatter plot from the actual data set.
The machine is taught to learn by importing the dataset into regression analysis model with the .fit
command and inserting the required values to predict the outcome for the model to analyze the data. In
the last step, all the values obtained from the forecast are used to create Prediction Line onto the Scatter
Plot that has been created. To compare the accuracy of the values obtained from the forecast with the
actual data, data scientists can use the Scikit-Learn library, by writing the program as presented in the
example.
From the graph above, the scientists create the feature height and the target variable weight. A Scat-
ter Plot is then created, displaying the data as a data point that intersects the x-axis which is the height,
and the weight in the y-axis. The LinearRegression() command from the Scikit-Learn library is run and
then the height and weight datasets are to teach machine learning.
Using supervised learning regression analysis, the scientists can predict the weight from a height of
175 cm. The result, 75.97 kg, is displayed via the print command. All forecast data are then used to cre-
ate the Prediction Line to show the closeness of the data obtained from the forecast with the actual data.

CLASSIFICATION

Data classification is a type of supervised learning among all data mining techniques in which data
scientists divide data into two parts: the data used to teach machines and the data used for testing. The
target variables are set as the result of data classification. The questions are consistent with techniques
used in data analysis, such as classifying the bank customers who have ability-to-repay if refinancing the
loan or classifying the coffee shop customers who show possibility to go to other coffee shops. There
are data mining techniques that are characterized as data classification techniques such as decision trees
(Cheewaprakobkit, 2019).

Data scientists can program for analyzing data with decision tree techniques as in the syntax below.
from sklearn.datasets import sample dataset
from sklearn import tree
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
Model = tree.DecisiontreeClassifier()
Feature using Train, Feature using Test, Target variables using Train, Target variables using Test =
train_test_split(sample dataset.data, sample dataset.target, test_size = 0.4)
model.fit(Feature using Train, Target data using Train)
Result variables from the actual test = Target variable using Test

277
 
Data Mining for Junior Data Scientists

Result variables from data classification= model.predict(Feature using Test)

Accuracy result = accuracy_score(Target variables using Test,Result variables from data classification)
print(accuracy result)
tree.plot_tree(model.fit(Feature using Train, target variables using Train))

Figure 46. Linear regression analysis

From the syntax, data scientist can use the sample dataset from sklearn.datasets to experiment with
programming for large data processing, then classify the data to import tree to run a set of .Decision-
TreeClassifier() command in order to classify data with decision trees. This is to divide the data into

278

Data Mining for Junior Data Scientists

datasets; for teaching machine learning and for testing. The train_test_split command from sklearn.
model_selection can be run for testing the accuracy of data analysis when the data scientists can use
accuracy_score command from sklearn.metrics.
While processing the data, scientists pack decision tree techniques into variables. Then, the sample
data set is divided into 2 sets. The 1st set is the features used to teach machine learning and target vari-
ables that teach machine learning, 60 percent of all instances. The 2nd set is the features used to test and
target variables, 40 percent of all instances.
The data are then categorized by the decision tree technique, importing the data for machine learning
into the model via .fit commands, and then classifying the data from the feature used for testing together
with the model being taught to learn through the .predict command. Then the efficiency of data clas-
sification is tested with the accuracy_score command, with the test target variables and with the data
classification. The results are then presented via the print command in the last step. The data scientists
can display the results of data classification with decision trees using the tree.plot.tree command. The
scientists can program by running the iris dataset, which stores attributes including the width of the pet-
als, the length of the petals, the width of the sepals and the length of the sepals, totaling 150 instances.
Then the program can be written to identify the Iris cultivars, including Setosa, Versicolor and Virginica,
which are assigned as target variables. The program is in the following example.

Figure 47. Programming for data analysis with decision trees

279
 
Data Mining for Junior Data Scientists

From the figure, data scientists have divided programming into 3 parts: 1. Developing visual data
in the Scatter Matrix for surveying the data, 2. Analysis of data using decision tree technique, and 3.
Accuracy display of data analysis and data classification results with decision trees.
Part 1: The data scientists retrieve the iris dataset from load_iris(), then create a data frame from
the iris dataset, and create a Scatter Matrix visualization through the .pairplot command by graphing it
from the features; the petal width, the petal length, the petal width and the sepal length, and the target
variables; the Iris flower species. The result of Execute is shown in the picture.

Figure 48. Scatter matrix visualization for data exploration

Part 2: The data scientists pack decision tree techniques into the model variable with tree.Decision-
TreeClassifiter command, and then divide the dataset used to teach machine learning which is iris.data and
the dataset used for testing which is irist.target into 2 parts, 60% for machines learning and 40 percent for
testing. The data in each set also includes the characteristic features of the Iris flowers and the cultivars.

280

Data Mining for Junior Data Scientists

The machine learning dataset is then imported into the model via .fit, and the data is categorized with
.predict command, processing the test features with the data, loaded into the model. The classification
accuracy is tested by accuracy_score command by using the target variable data for testing together
with the results obtained from the classification data. When executing the command set, the accuracy
of the model is obtained as follows:

Figure 49. Accuracy of data analysis with decision trees

From the figure, the accuracy of the data analysis is 91.66%, which means that the data obtained
from the classification has a high degree of accuracy and closeness to the facts.
In Part 3, the scientists build decision trees obtained from data analysis, using the tree.export_graphviz
command to create a graph from the data obtained from the analysis in Part 2. The graph can be used to
describe how Iris flower cultivars are classified based on the features and target variables.

REGRESSION

When the scientists are asked for a data analysis aimed at predicting quantitatively correlated data, such
as customer revenue and product sales, they can use regression analysis to analyze the relationship and
predict it. The variables used for data analysis can have one or more features for the analysis of target
variables (Labels). Therefore, regression is a type of data mining technique, Supervised-Learning, which
must be divided into 2 sets: data for the machine learning and data for model testing.
In addition, this book has already introduced programming for simple regression analysis. The discus-
sion of regression analysis in this section, therefore, covers programming for multiple linear regression
and polynomial regression.

1. Multiple Linear Regression

In this case, when the scientists intend to predict the outcome of the target variables using two or
more feature data, the multiple regression analysis is used (Rathaur, Kamath, & Ghanekar, 2020). The
programming syntax is as follows.

from pandas as dataset variables

import numpy as combined variables Feature
from sklearn.linear_model import LinearRegression
from sklearn.model_selection import train_test_split
dataset storage variables = dataset variables read_csv(‘data source’)
storage data Feature = dataset variables.DataFrame(combined variables Feature.c_[df[‘Feature’],
df[‘Feature’], columns = [‘Feature’, ‘Feature’])

281
 
Data Mining for Junior Data Scientists

target variables = df[‘Label’]

Feature using Train, Feature using Test, target variables using Train, target variables using Test =
train_test_split(storage variables Feature, target variables, test_size = 0.3)
Model variables = LinearRegression()
Model variables.fit(Feature using Train, target variables using Train)
Forecast results = model variables.predict(Feature using Test)
print(forecast results)
model variables.score(Feature using Test, target variables using Test)

Figure 50. Decision tree model

From the syntax that uses multiple regression analysis techniques to predict target variables with
two or more features, data scientists can experiment with programming. The sample program is divided
into 3 parts: 1. Feature Selection to select variables suitable for data analysis, 2. Graphing in 3D Scatter
Plot format to survey data from variables obtained from the section 1, and 3. Using multiple regression
analysis techniques to analyze the data of the target variables.
The goal of this analysis is to analyze which features influence daily rental pricing, and how much
the accommodation should be priced each day. The scientist can write a program as in this example.

282

Data Mining for Junior Data Scientists

Figure 51. Multiple regression analysis

As you can see, the data scientists write a program in three parts.
Part 1: Data scientists run the read_csv command from the pandas library to import the dataset into
the instruction set, and then perform a correlation analysis between all features and the target variable,
which is the rental price per day by using the .corr() command to analyze the top 3 correlated features.
As a result, price which is the target variable, and Feature which is longitude, the property’s location, and
availability_365 which is the number of days of service per year. This feature selection process is known
as feature selection. Data scientists can use this approach along with consulting knowledge experts, data
subjects, or questioners who intend to analyze them in order to obtain the features that are effective in
their performance. Then, the results of the data analysis are obtained as follows.

283
 
Data Mining for Junior Data Scientists

Figure 52. Feature selection operation

Part 2: After data scientists have acquired the features that can be used to predict target variables, the
scientists can take both features and target variables to survey the data to determine the distribution of
the data. To observe outlier data and show the relationship of data in 3D format that creates a Data Point
from data on the x, y and z axes with 3D Scatter Plot visualization data, data scientists can bring the
feature named availability_365 to build on the x-axis, and the longitude to build on the y-axis. Then, the
target variable, price, is generated on the z-axis, resulting from the development of the visualization data.

Figure 53. Surveying data with 3D scatter plot

284

Data Mining for Junior Data Scientists

Part 3: Data Analysis with Multiple Regression Analysis

Data scientists take features named availability_365 and longitude to create a data frame, and populate
into the observation variable x. To program for data analysis with multiple regression analysis techniques,
it uses a method of combining features by creating Data Frame before analyzing the data together with
the target variables.
The target variable, price, is then inserted into the y variable, and the data within the variables x and
y are divided into two parts: Part 1 is 60% for machine learning, and Part 2 is 40% for model testing.
Then the model variable is assigned to used LinearRegression(), and the machine is taught to learn by
applying a Train-based feature and a Train-based target variable into the model.
Data scientists forecast daily accommodation prices. The .predict command is used in conjunction
with the Test-based feature, and all forecasts are rendered via the print command, and the precision of
the data is analyzed in the R Squared format. The result is 0.0274, which is still very accurate. Accuracy
is considered a low level, which means that the data received from the forecast is not closely related to
the facts as shown in the figure.

Figure 54. Forecast results and model accuracy values

To analyze data using multiple regression analysis techniques, Python uses the LinearRegression()
command similarly to a simple regression analysis. Only the method of combining multiple features to
create a data frame is done before analyzing with the target variables.

2. Polynomial Regression

In the case of data scientists survey the data through the creation of a scatter plot, and the data is not
linear, data scientists can use polynomial regression analysis techniques for predictions which is more
closely tied to real data (Gong & Zhang, 2021). The syntax is as follows.

import matplotlib.pyplot as graph making variables

from sklearn.linear_model import LinearRegression
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import PolynomialFeatures
polynomial variables= PolynomialFeatures(degree = number of Matrix)
Feature variables= [data]
Target variables = [data]
Polynomial storage variables = polynomial variables.fit_transform(Feature variables)
print(Polynomial storage variables)
Model variables = LinearRegression()
Model variables.fit(Polynomial storage variables, target variables)

285
 
Data Mining for Junior Data Scientists

Forecast results = model variables.predict(Polynomial storage variables)

print(forecast result)
graph making variables.scatter(Feature variables,target variables)
graph making variables.plot(Feature variables, forecast result)
model variables.score(Polynomial storage variables, target variables)

From the syntax, data scientists can define a matrix used to adjust a Data Object within a Feature
variable by setting the number of Degrees for PolynomialFeatures, then use the .fit_transform command
to transforms the Data Object into a Matrix form in order to bring the values ​to forecast the target vari-
able’s value. Therefore, they can create a graph showing the intersection of the actual data together with
the Prediction Line obtained from the forecast with high accuracy. Data scientists can write a program
as follow.

Figure 55. Programming with polynomial regression analysis technique

In the figure, the data scientists transform the normal data into the Matrix form with the Polynomail-
Features command, based on the equation Y = B0+B1X+B2X2. In the example, Degree is set as 2,
which means the matrix with the number 0. – 2 sums to 3, and then the values inside the variable h are
converted into such forms before putting into the variable h_poly. The result of the data transformation
is as follows:

286

Data Mining for Junior Data Scientists

Figure 56. Results of data processing with polynomial features

Data scientists combine the h_poly variable, which acts as a feature, together with the variable w,
which acts as a target variable. The machine is taught to learn through the .fit command. Data is then
analyzed with LinearRegression() command then predicted with the variable h_poly. The forecast result
is presented via the print command.

Figure 57. Forecast results with polynomial regression analysis

To show the efficiency of data analysis, data scientists create a scatter plot based on the data of the
initial variables h and w. A Prediction Line is then created from the data obtained from the forecast
together with the variable h_poly which is generated from the data within the variable h transformed
by PolynomailFetures to demonstrate the affinity of facts. The .score command is used to analyze the
accuracy. The result is 94.74% which is high accuracy, as shown in the figure.
Polynomial regression analysis is the transformation of data into matrix format, which is suitable for
using non-linear features to analyze target variable data with LinearRegression() command. According
to the samples, the forecast results show high accuracy.

FEATURE EXTRACTION

Data mining techniques use computational methods to process datasets in order to obtain results that can
be used to answer a given question. Therefore, the data that can be used in data mining techniques must
be numerical. However, data scientists often receive datasets that are either text or images, which cannot
be imported into data mining techniques directly. Data scientists therefore need to use Feature Extrac-
tion to transform data from text or images into a numeric form so that it can be used for computation
during the data preparation process. The library Scikit-Learn provides the datasets that data scientists
can apply for Feature Extraction

287
 
Data Mining for Junior Data Scientists

Figure 58. Results of data analysis by polynomial regression analysis technique

1. Sklearn.Feature_Extraction

The Scikit-Learn library provides a package named Sklearn.Feature_Extraction which stores a set of
commands for feature extraction to convert text into numbers in an array format as in the following syntax:

from sklearn.feature_extraction import DictVectorizer

data processing variables = DictVectorizer
dataset variables = text data
data processing variables. fit_transform(data variables).toarray()
data processing variables. get_feature_name()

From the syntax, data scientists run the DictVectorizer command from skleawrn.feature_extraction,
create variables that perform data processing using DicVectorizer command, and then create a text data set.
Data scientists can use .fit_transform command in combination with the dataset and the .toaary()
command to convert all text into numbers in the form of an array, and display the feature names within
the dataset with the .get_feature_name() command. You can write a program for Feature Extraction as
in the example.
In the figure, data scientists determine the vec variable to manipulate the data using DicVectorizer()
command, then create a dataset consisting of text and numbers, pack them into variable m, and con-
vert them into numbers as an array by taking the variable m to process in the .fit_transform().toaary()
command. And the result of the Execute command is a whole numeric array. The feature name is then
displayed with the get_feature_names command. The result of the execution is the text Covid=19, and
the list of provinces includes Bangkok, Chiang Mai and Lampang, which is the text data.

288

Data Mining for Junior Data Scientists

Figure 59. Using the DicVectorizer command

2. Sklearn.Featue_Extraction.Text

Where data scientists receive long text or sentence datasets and intend to transform them into a text,
they can run CountVectorizer command from package sklearn.feature_extraction.text, as in the follow-
ing syntax.

from sklearn.feature_extraction.text import CountVectorizer

dataset variables = [‘data’]
data processing variables = CountVectorize(analyzer= ‘word’)
data processing variables. fit_transform(dataset variables).toarray()
data processing variables.get_feature_names()

From the syntax, data scientists run CountVectorizer command from sklearn.faeture_extraction.text pack-
age and then create a variable that transforms text into numeric data using the CountVectorizer(analyzer=
‘word’) command. Then the variables are used to process the data within the dataset with the .fit_transform().
toarry() command, and the feature names within the dataset are displayed with the .get_feature_name()
command. Data scientists can write the syntax as in the example.
In the figure, the scientists run CountVectorizer(analyzer= ‘word’) command to define vec variable to
process the data. The variable that stores the dataset is then inserted into .fit_transform.toarray() command
in order to convert text into numeric data in an array format. The .get_featuer_names() command is then
used to present feature names from the source dataset, and then displayed through the print command.

289
 
Data Mining for Junior Data Scientists

Figure 60. Using CounVectorizer command

3. Feature Hashing

In case data scientists intend to determine the number of Data Objects converted from text into numeric
data within the sklearn.feature_extraction package, the FeatureHasher command is available to use.

from sklearn.feature_extraction import FeatureHasher

data transforming variables = FeatureHashwer(n_features=number of Data Object)
dataset variables = [{data}]
data after transformation = data transforming variables.transform(dataset variables)
data after transformation. toarray()

From the syntax, the scientists run FeatureHasher command from sklearn.feature_extraction pack-
age by specifying the number of Data Objects after processing, then use the .transform command to
transform the data, and then display it in the form of an array using the .toarray() command.

Figure 61. Using FeatureHasher

290

Data Mining for Junior Data Scientists

From the figure, data scientists runs FeatureHasher command from sklean.feature_extraction package
by defining the hash variable to process the data and assigning n_feature as value of 10. After processing
the data, there will be 10 numeric Data Objects. Then the dataset is packed into Animal variable, and
the data is transformed into numeric data with .transform() command. The data is then packed into the
data variable and displayed as an array with .toarray() command.

CONCLUSION

When data scientists begin to study Python for big data analysis, they first need to realize that Python
provides a number of libraries that can be used to support data analysis, including Pandas, NumPy, SciPy,
and Scikit-Learn. Besides studying about the commands and programming methods, knowledge of data
mining is also something that data scientists need to explore in order to be able to choose the right data
mining technique for the given problem. Python also supports data exploration through visualization and
data preparation. Python is a tool that supports data science processes and big data analytics as it reduces
the use of digital resources. In addition, this book also uses Google Colab tools where data scientists can
write and test Python commands through a Web Brower without the need to install the Software Editor
and Interpreter for libraries on their computers.

EXERCISE QUESTIONS

1. Write a program to create a data frame from a data set like this:
2. Write a program to create a data frame from a data consists of 16, 20 and NA then compensate NA
by mean of 16 and 20.
3. Write a program with one-hot encoding from Table 1.

Table 1.

Colors
Red
Red
Yellow
Green
Pink

4. Explain the difference in data analysis between Supervised-Learning with Unsupervised-Learning

5. From the dataset includes mileage and oil. Apply polynomial regression analysis to predict oil from
distance.

Distance [2.5,4.6,5.6,3.2,6.6,3.1,3.2,3.3,4.5]
Oil [2.2,5.5,5.2,4.1,4.2,4.4,3.2,2.2,1.4]

291
 
Data Mining for Junior Data Scientists

REFERENCES

Cheewaprakobkit, P. (2019). Predicting Student Academic Achievement by Using the Decision Tree and
Neural Network Techniques. Asia-Pacific International University.
Gong, Y., & Zhang, P. (2021). Predictive Analysis and Research Of Python Usage Rate Based on Poly-
nomial Regression Model. 2021 3rd International Conference on Artificial Intelligence and Advanced
Manufacture (AIAM), Artificial Intelligence and Advanced Manufacture (AIAM), 2021 3rd International
Conference on, AIAM, 266–270.
Healy, K. (2018). Data visualization: a practical introduction. Princeton University Press.
Massaron, L., & Mueller, J. (2015). Python for data science for dummies. John Wiley and Sons, Inc.
Mueller, J., & Massaron, L. (2019). Python for data science (2nd ed.). John Wiley & Sons, Inc.
Pajankar, A. (2021). Practical Python data visualization: a fast track approach to learning data visual-
ization with Python. Apress. doi:10.1007/978-1-4842-6455-3
Rathaur, S., Kamath, N., & Ghanekar, U. (2020). Software Defect Density Prediction based on Multiple
Linear Regression. 2020 Second International Conference on Inventive Research in Computing Applica-
tions (ICIRCA), Inventive Research in Computing Applications (ICIRCA), 2020 Second International
Conference On, 434–439.
Sonu, S. B., & Suyampulingam, A. (2021). Linear Regression Based Air Quality Data Analysis and
Prediction using Python. 2021 IEEE Madras Section Conference (MASCON), Madras Section Confer-
ence (MASCON), 1–7.

292
 293

Glossary

Administrator: Internal executives who establish the questions that data scientists use for data min-
ing analysis.

Agglomerative Clustering Algorithm: A clustering method that supports Hierarchical Clustering

by analyzing data from 2 or more attributes.

Apriori Algorithm: An algorithm used for analyzing correlation rules, using Confident values,
Support values, and Support and Confidence Thresholds values as a measure for generating correlation
rules in the form of Frequency Itemset.

Artificial Neural Network or ANN: A data mining technique that has both classification and nu-
merical prediction capabilities. It can be classified as Supervised Learning.

Association Rule: Rules for the relationship of data or events. When one event occurs, another event
will always follow.

Association Rule Mining: An analysis of the relationship rules of data or events from a dataset
related to that data or event.

Attribute: The data attributes such as gender, education, income.

Attribute Transformation: Changing data formats to support data analysis with data mining techniques.

Authentication: Authentication of a user in any form, such as authentication with User Name or
Password.

Average Pooling: Taking all the numbers in each part of the image data to create an average or mean.

Aviation Function: Adjusting data from Linear to Non-Linear with values between 0 - 1, which
makes the data obtained from the calculation more closely attached to the facts until getting the output
from calculation.



 Glossary

Back Propagation Artificial Neural Network (BP-ANN): Teaching machines to learn and test data
with a neural network, which focuses on adjusting the weights of each link.

Bayesian Classifier: Classification of data based on the probability that data will be classified into
a particular type.

Bias: A careful selection of data analysis to reduce the tendency of the data from the concentrated
opinions.

Big Data: Big data can grow from customer transactions connected over the Internet until it acquires
a large dataset that hides customer behavior patterns.

Binominal: Data Characteristics within RapidMiner with values of 0 and 1.

Box Plot: A graph showing the lowest value, 25% value, middle value, 75% value, highest value and
abnormal data as well as showing multiple groups of data simultaneously.

Branch: Results in data classification within the Decision Tree.

Business Understanding: Understanding the issues of each business, the business environment and
context of the industry.

C-Index: Clustering performance analysis that can avoid discrepancies from data points, character-
ized by noise.

Categorial Data: Data Categories such as gender, education, flower species.

Centroid: The center of the clustered data.

Class Label: Identifying the types of data that can be classified within the decision tree.

Classification: Data mining techniques used to classify data such as Decision Tree or k-NN.

Classification Rule: Classification is possible only if there is a result of a variable that supports the
classification efficiency.

Cloud Technology: Technology to provide software services, processing, and memory without
installing any hardware or software.

Cluster: Data groups such as groups of customers with different purchasing behaviors.

294
Glossary

Clustering: Grouping the data, segmenting customers from the available data sets in order to know
how differently each customer behaves.

Collaborative Filtering or CF: The focus on analyzing the behavior of customers with similar interests.

Confidence: The frequency as a percentage of the availability of rules-based data within the dataset.

Confusion Matrix: Measurement of the efficiency of data classification.

Content-Based Filtering: The focus on user behavior analysis.

Content-Based Recommendation: The focus on recommendations based on past content as references.

Continuous Data: Data that shows actual numbers such as height, distance, salary.

Convolutional Neural Networks: One of the deep learning techniques used to analyze large amounts
of data, especially image data.

Convolutional Operation: Image data analysis in numeric form.

CRISP-DM: A process that enables data scientists to understand the problem and choose the right
data mining techniques, which leads to the correct answer to the determined questions.

Cross Validation: Determination of the number of data division cycles by dividing the data into 2
parts in every round.

Data Analytics: The use of the Data Mining Technique for analyzing big data from the business
sectors or other industries through classifying, forecasting or analyzing correlation rules to answer de-
termined questions.

Data Collection: Collecting or even extracting data from outside the organization such as Social
Network.

Data Exploration: Exploring data patterns and trends as well as surveying unusual and incomplete
data.

Data Labeled: Target variables that data scientists need for forecasting or classifying data from the
existing datasets, where the target variable is one of the components of the Supervised-Learning processing.

Data Mining: The techniques for analyzing data from large datasets, divided into Supervised-Learning
and Unsupervised-Learning. The results obtained from the analysis are in the form of data classification,
data segmentation, data forecasting and correlation rules.

295
 Glossary

Data Object: The data within the attributes of a dataset.

Data Point: The intersection that occurs when the data of 2 Attributes are plotted on the Scatter Plot.

Data Pre-Processing: Changing the data format to be standardized until the data can be processed
by data mining techniques.

Data Preparation: The preparation of data by dealing with outliers and incomplete data.

Data Science: Knowledge on algorithms development to analyze data from large datasets. The result
of development is in the form of Data Mining.

Data Scientist: The person who analyzes problems, explore and prepare data, analyze data with data
mining techniques and present the results obtained from the analysis.

Data Selection: A selection of data to support the analysis of data with data mining techniques, and
to answer the determined questions.

Data Set: The data set locates inside and outside the organization. It is consistent with data analysis.

Data Sources: A source of data sets that can be stored either internally or externally, such as a busi-
ness partner’s data store.

Data Store: To store data, data scientists should duplicate the data used for analysis without altering
the original data.

Data Understanding: Understanding the nature of the data whether it can be used to analyze the
data to answer the determined questions.

Data Visualization: Creating graphs to understand data patterns, data trends and abnormal data
during the survey process.

Date: Data characteristics within RapidMiner covering the information on day, month, year.

Date_time: Data characteristics within RapidMiner which include day, month, year, and duration.

Davies-Bouldin Index: An analysis of the performance of clusters using data from the center of
every cluster (Centroid).

DBSCAN: Clustering data by defining the radius from the center from every Data Point and setting
the minimum number of Data Points that must be within the radius in order to determine the density of
data within each cluster.

296
Glossary

Decision Tree: A technique to classify the resulting data in a tree diagram. Data scientists can use
the diagram to classify data or target variables.

Deep Learning: A data mining technique that can be used to analyze non-projected data; including
images, sounds, and text. There are two types of deep learning; convolutional neural networks and re-
current neural networks. The deep learning model modulates the pre-configuration data for processing
in conjunction with neural network techniques.

Deployment: The use of the results obtained from the analysis to respond to the problem.

Descriptive Statistics: The statistics that focuses on quantitative analysis of data. It could be an
analysis of the data to answer a given question, such as testing the average between two populations.

Dimensionality: The number of attributes within the data set.

Discreate: The data that indicates a clustering or a data type.

Domain Expert: A data specialist who is familiar with data sets that can be used to analyze problems.
Data scientists must consult with the expert to understand the data and ensure that the results obtained
from the analysis are consistent with the corporate environment.

Dunn Index: The determination of the minimum distance from the data point to the data point in
other clusters (Intercluster) divided by the distance of the largest data point within the cluster.

Duplicate Data: Duplicate data within the same data set.

Entropy: The uncertainty of the data.

Epochs: Iterative analysis of the data to teach the machine to learn each time.

Euclidean Distance: The Calculation of the distance between the data point and the center of each
cluster within a K-Mean clustering. Within the data set, there is still the movement of the data object
while clustering the data in multiple direction.

Feature: Features in database textbooks is also called an Attribute.

Feature Extraction: Image transformation process to be in numeric format.

Feature Selection: Selecting appropriate features to be used in data analysis, using quantitative
analysis techniques or consulting a Domain Expert.

297
 Glossary

Filter Matrix: It is used to examine patterns in each part of the image to obtain data within the feature
that preserves the distinctive characteristics of the image.

Forecasting: Numerical forecasting of outcomes. In some data mining techniques, such as logistic
regression, it may result in data classification.

Frequency: The frequency at which an event occurs or similar information is encountered.

Fully Connected: The merging of data segments in a feature that is ready to be used to process data
with neural networks for classifying image data.

Generalized: Organizing data in a format suitable for classification.

Hashtag: Text used to characterize content within Twitter, preceded by the # sign.

Hidden Layer: Nodes and layers that enhance neural network processing efficiency.

Hierarchical Clustering: The division of data into cluster according to the order of clusters to de-
termine which sub-clusters belong to each large cluster.

Histogram: A graph used to explore the distribution of data by generating the spectral values of the
data.

Impurity Measure: A value at which each class can be divided into more classes. The better the
class can be divided, the lower the impurity measure.

Inconsistency Data: The data gathered from multiple but conflicting sources.

Inferential Statistics: A statistical analysis once receiving the results of a test on a sample. It can
be used to represent the entire population.

Information Gain: Selecting the features appropriate for creating a decision tree.

Information Security: Information security, which includes data integrity, data confidentiality, and
data availability. This can be done both in terms of security policy and the provision of effective defense
technology.

298
Glossary

Input Neural: The number of input nodes dependent on the number of variables or features analyzed
with the model.

Instance: Number of records formed by a combination of multiple features.

Integer: Data attributes within RapidMiner which are integers.

Interpretation: Translating the results of data mining analysis into an easy-to-understand format.

Item-Based Filtering: It focuses on analyzing product properties.

Item-Based Nearest Neighbor Recommendation: An analysis on similarity of a dataset, which is

a human opinion and the similarity of the properties of the subject.

Itemset: The number of features to be analyzed in the correlation rule.

K-Means: Clustering and defining according to the objectives of the data analysis. It is a type of
Unsupervised-Learning datamining techniques.

Knowledge Discovery: The discovery of knowledge in various forms, such as the use of data mining
techniques to analyze the correlation rules of customer buying behavior.

Knowledge Discovery Process: The process of analyzing data using data mining techniques, focus-
ing on data extraction, data processing using data mining and interpretation techniques to answer the
questions.

Label: Target variables that are set for forecasting or classifying data.

Leaf Node: The result of classifying data from each class within the decision tree.

Long Short-Term Memory (LSTM): The ability to memorize values from the previous step that
are very distant from the current node. Those values are then computed with the current node to obtain
results from sorting related data within deep learning data mining techniques.

Machine Learning: A machine that has the ability to recognize characteristics of data for data fore-
casting, data classification, data segmentation and correlation rule analysis.

299
 Glossary

Manhattan Distance: A method for calculating the distance for quantitative data by subtracting the
data objects from each instance and gathering the results of all data objects.

Max Pooling: Selecting the largest number in each part of the image data.

Missing Value: Incomplete data within the dataset.

Mode: Frequency analysis from the largest amount of duplicate data.

Model Evaluation: Analysis of efficiency to assure how accurately data processing can predict,
classify, and analyze correlation rules.

Modeling: Data analysis with data mining techniques without programming. The tools within software
packages such as RapidMiner can be used to create a schematic data analysis process.

Naïve Bayes: Classification of data using probabilistic analysis as to what classification the data
should be.

Nearest Neighbor: Matching the closest neighbors by identifying how the data is classified from
the neighboring data.

Nominal Scale: Data in the form of clusters, such as gender; female, male and others.

Normalization: Making data to the same standard.

Numerical Prediction: Forecasting numeric data by defining target variables used for forecasting.

Objective Measurement: Using data from quantitative analysis that can be used to reflect the ef-
ficiency of correlation rules obtained from data mining analysis.

Open Source: Non-licensed software that allows developers from all over the world to collaborate
on the development of the software.

Ordinal Scale: Representing an Abstraction with ordinal numbers.

Outlier: Outliers, either too many or too few, can affect the average of the entire feature. It may be
due to data entry errors.

Output Neural: The results obtained from the processing. There can be one or more results depend-
ing on the data analysis question.

300
Glossary

Pattern Recognition: A system for recognizing and distinguishing data from the learned patterns,
such as photo recognition.

Pearson Correlation: Analysis of the relationship between two sets of data for similarity analysis.

Polynomial: Characteristics of data within RapidMiner that can be numeric data or text data.

Polynomial Regression: Data analysis for forecasting. Data obtained from this analysis is more
closely related to facts than linear data analysis.

Precision: Displays of the accuracy of data classification.

Predict Accuracy: The difference in data that data scientists use to teach machines to learn.

Prediction: Results from data analysis using data mining techniques to predict the outcomes based
on data from existing datasets.

Probabilistic Method: One of the methods for forecasting missing data using Bayes Theorem.

Pruning the Tree: The reduction of the feature number used in data analysis in order to ease its use
in data classification.

Rank Accuracy: The percentage accuracy of the recommendation that Recommendation System
provides for users.

RapidMiner: The software that allows data scientists to process data with data mining techniques
by creating schematic models.

Real: The data within RapidMiner as real numbers.

Recall: The displays of the accuracy of data classification in accordance with the actual data.

Recommendation System: The system that analyzes data from a dataset. It can be used to support
decision-making in a variety of matters.

Recurrent Neural Networks: One of the deep learning data mining techniques aiming at analyzing
the relationships of sequential and time series data.

Regression: A data mining technique used to forecast numerical data.

301
 Glossary

Resolution: A measure used to store data objects in each feature.

Retweet: Repeating the Tweets that others have Tweeted in their Twitter Account.

Rule Interestingness: Determining the strength or value of correlation rules, or applying the results
of the analysis to make decisions in businesses or industries.

Scalability: The size of data that keeps growing from updates of global users on social networks or
data that grows from the updates from business partners.

Scatter Plot: A graph that creates intersections from 2 or more features by plotting graphs on the x
and y axes.

Set Role: Assigning a feature within RapidMiner software as a target variable for Supervised-Learning
datamining process.

Similarity: Calculation of similarity.

Split Data: Splitting data into two parts: 1 for teaching machine learning and 2 for testing the per-
formance of the model.

Stopping Conditions: The neural network model will stop computation in order to teach the machine
learning. Data scientists can then calculate weight to obtain the appropriate error level.

Stride: Moving the Filter Matrix to inspect the image data one by one.

Strong Association Rule: A strong correlation rule, used as a rule for data analysis. It is useful for
making decision.

Subjective Measurement: Using opinions or Domain Expert suggestions to decide which correla-
tion rules are helpful for decision making.

Supervised-Learning: Teaching machines to have the ability to recognize the features to classify,
forecast, and analyze the correlation rules.

Support: The frequency as a percentage of the availability of rules-based data within the entire dataset.

Support and Confidence Thresholds: A measure of whether each correlation rule can be used in
data analysis.

302
Glossary

Tensor Object: The model used to transform image data into numeric format.

Test Set: A data sets for testing the performance of a data mining model.

Time: A characteristic of data within RapidMiner, characterized by duration.

Time Series: Data that occurs continuously over time.

Training Set: A data set for teaching machines to learn.

Tweet: Commenting or presenting products and services within Twitter.

Twitter: A social network where users can comment and reproduce the opinions of others.

Twitter API: A tool used for extracting data from Twitter for the purpose of data analysis.

Unsupervised-Learning: Analyzing data with data mining techniques, that do not require machine
learning. Those datasets can be processed directly, such as Cluster Analysis and Association Rule Mining.

User-Based Nearest-Neighbor Collaborative Filtering: Forecasting current missing data based on

other data of a similar and complete form.

Web Mining: Data Extraction used for data analysis from web-based resources to obtain a dataset
that corresponds to the questions.

303
304

Compilation of References

Abbas, A., Prayitno, P., Butarbutar, F., Nurkim, N., Prumanto, D., Dewadi, F. M., Hidayati, N., & Windarto, A. P. (2021).
Implementation of clustering unsupervised learning using K-means mapping techniques. IOP Conference Series. Materi-
als Science and Engineering, 1088(1), 012004. doi:10.1088/1757-899X/1088/1/012004

Abbasnasab Sardareh, S., Brown, G. T. L., & Denny, P. (2021). Comparing Four Contemporary Statistical Software
Tools for Introductory Data Science and Statistics in the Social Sciences. Teaching Statistics: An International Journal
for Teachers, 43.

Abdul Aziz, A., & Starkey, A. (2020). Predicting Supervise Machine Learning Performances for Sentiment Analysis Us-
ing Contextual-Based Approaches. IEEE Access : Practical Innovations, Open Solutions, 8, 17722–17733. doi:10.1109/
ACCESS.2019.2958702

Abu-Soud, S. M. (2019). A Novel Approach for Dealing with Missing Values in Machine Learning Datasets with Discrete
Values. 2019 International Conference on Computer and Information Sciences (ICCIS), 1-5. 10.1109/ICCISci.2019.8716430

Adeniyi, D. A., Wei, Z., & Yongquan, Y. (2016). Automated web usage data mining and recommendation system using
K-Nearest Neighbor (KNN) classification method. Applied Computing & Informatics, 12(1), 90–108. doi:10.1016/j.
aci.2014.10.001

Aditya, G. M., Hoode, A., Anvesh Rai, K., Biradar, G., Ajay Kumara, M., Manoj Kumar, M. V., Prashanth, B. S., Sneha,
H. R., & Shivadarshan, S. L. (2021). Machine Learning Based Platform and Recommendation System for Food Order-
ing Services within Premises. 2021 2nd Global Conference for Advancement in Technology (GCAT), Advancement in
Technology (GCAT), 2021 2nd Global Conference For, 1–8.

Aggarwal, C. C. (2018). Neural networks and deep learning: a textbook. Springer. doi:10.1007/978-3-319-94463-0

Almasoud, A. M., Al-Khalifa, H. S., & Al-Salman, A. S. (2019). Handling Big Data Scalability in Biological Domain
Using Parallel and Distributed Processing: A Case of Three Biological Semantic Similarity Measures. BioMed Research
International, 2019, 1–20. doi:10.1155/2019/6750296 PMID:30809545

Andrews, F. M., & Messenger, R. C. (1973). Multivariate nominal scale analysis; a report on a new analysis technique
and a computer program. University of Michigan.

Angiulli, F., & Fassetti, F. (2021). Uncertain distance-based outlier detection with arbitrarily shaped data objects. Journal
of Intelligent Information Systems, 57(1), 1–24. doi:10.100710844-020-00624-7

Atanasovski, M., Kostov, M., Arapinoski, B., & Spirovski, M. (2020). K-Nearest Neighbor Regression for Forecasting
Electricity Demand. 2020 55th International Scientific Conference on Information, Communication and Energy Systems
and Technologies (ICEST), Information, Communication and Energy Systems and Technologies (ICEST), 2020 55th
International Scientific Conference On, 110–113.



Compilation of References

Baby, S. S., & Reddy, S. L. (2021). End to End Product Recommendation system with improvements in Apriori Algorithm.
2021 Third International Conference on Inventive Research in Computing Applications (ICIRCA), Inventive Research
in Computing Applications (ICIRCA), 2021 Third International Conference On, 1357–1361.

Bader, D. (2018). Python tricks: The book. Dan Bader.

Başarslan, M. S., & Argun, İ. D. (2018). Classification Of a bank data set on various data mining platforms. 2018 Electric
Electronics, Computer Science, Biomedical Engineerings’ Meeting, 2018, 1–4.

Bayram, E., & Nabiyev, V. (2020). Image segmentation by using K-means clustering algorithm in Euclidean and Ma-
halanobis distance calculation in camouflage images. 2020 28th Signal Processing and Communications Applications
Conference (SIU), Signal Processing and Communications Applications Conference (SIU), 2020 28th, 1–4.

Bean, R. (2022). Why Becoming a Data-Driven Organization Is So Hard. Harvard Business Review Digital Articles, 1–6.

Bell, J. (2015). Machine learning: hands-on for developers and technical professionals. John Wiley & Sons.

Ben Ncir, C.-E., Hamza, A., & Bouaguel, W. (2021). Parallel and scalable Dunn Index for the validation of big data
clusters. Parallel Computing, 102, 102. doi:10.1016/j.parco.2021.102751

Beohar, D., & Rasool, A. (2021). Handwritten Digit Recognition of MNIST dataset using Deep Learning state-of-the-art
Artificial Neural Network (ANN) and Convolutional Neural Network (CNN). 2021 International Conference on Emerg-
ing Smart Computing and Informatics (ESCI), Emerging Smart Computing and Informatics (ESCI), 2021 International
Conference On, 542–548.

Berman, J. J. (2018). Principles and practice of big data: preparing, sharing, and analyzing complex information (2nd
ed.). Academic Press/Elsevier.

Bing, L., Yiming, M., & Lee, R. (2001). Analyzing the interestingness of association rules from the temporal dimension.
Proceedings 2001 IEEE International Conference on Data Mining, Data Mining, 2001. ICDM 2001, Proceedings IEEE
International Conference on, Data Mining, 377–384. 10.1109/ICDM.2001.989542

Bowles, M. (2015). Machine learning in Python: essential techniques for predictive analysis. Wiley.
doi:10.1002/9781119183600

Bramer, M. (2016). Principle of Data Mining (3rd ed.). Springer-Verlag London Ltd.10.1007/978-1-4471-7307-6

Cai, S., Sun, R., Hao, S., Li, S., & Yuan, G. (2016). An efficient outlier detection approach on weighted data stream
based on minimal rare pattern mining. China Communications, 16(10), 83-99.

Carlson, K., Livermore, M. A., & Rockmore, D. N. (2020). The Problem of Data Bias in the Pool of Published U.S.
Appellate Court Opinions. Journal of Empirical Legal Studies, 17(2), 224–261. doi:10.1111/jels.12253

Chaimongkol, N., & Meesad, P. (2010). Association Rule Mining for Specific New Course. Mahanakorn University of
Technology (MUT).

Chavan, A. (2020). RapidMiner Installation on Windows. Retrieved 15 December 2021 from https://www.youtube.com/
watch?v=4RpepmDH-7s

Cheewaprakobkit, P. (2019). Predicting Student Academic Achievement by Using the Decision Tree and Neural Network
Techniques. Asia-Pacific International University.

305
 Compilation of References

Chen, Q., & Qin, J. (2021). Research and implementation of movie recommendation system based on deep learning.
2021 IEEE International Conference on Computer Science, Electronic Information Engineering and Intelligent Control
Technology (CEI), Computer Science, Electronic Information Engineering and Intelligent Control Technology (CEI),
2021 IEEE International Conference On, 225–228.

Chiarini, T. M., Kohli, R., & Forsgren, N. (2021). Theories in Flux: Reimagining Theory Building in the Age of Machine
Learning. Management Information Systems Quarterly, 45(1), 455–459.

Chinnarasri, C., Nonsawang, S., & Supharatid, S. (2012). Application of Artificial Neural Networks for River Stage
Forecasting in Hatyai. King Mongkut’s University of Technology Thonburi. KMUTT.

Chisholm, A. (2013). Exploring Data with RapidMiner: Explore, Understand, and Prepare Real Data Using Rapidminer’s
Practical Tips and Tricks. Packt Publishing.

Chow, T. W. S., & Cho, S.-Y. (2007). Neural networks and computing: learning algorithms and applications. Imperial
College Press. doi:10.1142/p487

Curry, E., Heintz, F., Irgens, M., Smeulders, A. W. M., & Strmigioli, S. (2022). Partnership on AI, Data, and Robotics.
Communications of the ACM, 65(4), 54–55. doi:10.1145/3513000

Dai, H., Zhang, S., Wang, L., & Ding, Y. (2016). Research and implementation of big data preprocessing system based
on Hadoop. 2016 IEEE International Conference on Big Data Analysis (ICBDA), 1-5. 10.1109/ICBDA.2016.7509802

Desale, K. S., Kumathekar, C. N., & Chavan, A. P. (2015) Efficient Intrusion Detection System Using Stream Data Min-
ing Classification Technique. 2015 International Conference on Computing Communication Control and Automation,
469-473. 10.1109/ICCUBEA.2015.98

Emon, M. I., Shahiduzzaman, M., Rakib, M. R. H., Shathee, M. S. A., Saha, S., Kamran, M. N., & Fahim, J. H. (2021).
Profile Based Course Recommendation System Using Association Rule Mining and Collaborative Filtering. 2021
International Conference on Science & Contemporary Technologies (ICSCT), Science & Contemporary Technologies
(ICSCT), 2021 International Conference On, 1–5.

Erdoğmuş, Z. İ., & Arslan, M. K. (2022). Engaging with Social Media Influencers on Youtube: A Cluster Analysis.
Istanbul University Journal of the School of Business Administration, 51(1), 359–373.

Fadlil, A., Umar, R., & Gustina, S. (2019). Mushroom Images Identification Using Order Statistics Feature Extraction
with Artificial Neural Network Classification Technique. Journal of Physics: Conference Series, 1373(1), 012037.
doi:10.1088/1742-6596/1373/1/012037

Fernando, T., Gammulle, H., Denman, S., Sridharan, S., & Fookes, C. (2021). Deep Learning for Medical Anomaly
Detection - A Survey. ACM Computing Surveys, 54(7), 1–37. doi:10.1145/3464423

Ghorpade, V. G., & Koneru, V. S. (2021). Pattern recognition neural networkmodel for experimental based compres-
sive strength graded self compacting concrete. Materials Today: Proceedings, 43(Part 2), 795–799. doi:10.1016/j.
matpr.2020.06.175

Ghous, H., & Kovács, L. (2020). Efficiency Comparison of Python and Rapidminer. Multidiszciplináris Tudományok,
10(3), 212–220. doi:10.35925/j.multi.2020.3.26

Ginting, S. W., Hartati, R. S., Sudarma, M., & Swamardika, I. B. A. (2021). Clustering of Earthquake and Volcanic
Eruption Trauma Survivor Groups using K-Means Algorithm. 2021 International Conference on Smart-Green Technol-
ogy in Electrical and Information Systems (ICSGTEIS), Smart-Green Technology in Electrical and Information Systems
(ICSGTEIS), 2021 International Conference On, 69–73.

306
Compilation of References

Giri, C., & Chen, Y. (2022). Deep Learning for Demand Forecasting in the Fashion and Apparel Retail Industry. Fore-
casting, 4(2), 565–581. doi:10.3390/forecast4020031

Glass, D. H. (2013). Confirmation measures of association rule interestingness. Knowledge-Based Systems, 44, 65–77.
doi:10.1016/j.knosys.2013.01.021

Global Mindset University. (2020). RapidMiner Data Exploration demo | Data Analysis - Using the Iris Data | Data
Mining. Retrieved 15 December 2021 from https://www.youtube.com/watch?v=EfQkqcj3MIk

Gong, S., Ye, H.-W., & Zhu, X. (2009). Item-based collaborative filtering recommendation using self-organizing map.
2009 Chinese Control and Decision Conference, Control and Decision Conference, 2009. CCDC ’09. Chinese, 4029–4031.

Gong, Y., & Zhang, P. (2021). Predictive Analysis and Research Of Python Usage Rate Based on Polynomial Regres-
sion Model. 2021 3rd International Conference on Artificial Intelligence and Advanced Manufacture (AIAM), Artificial
Intelligence and Advanced Manufacture (AIAM), 2021 3rd International Conference on, AIAM, 266–270.

Görtler, J., Hohman, F., Moritz, D., Wongsuphasawat, K., Ren, D., Nair, R., Kirchner, M., & Patel, K. (2021). Neo:
Generalizing Confusion Matrix Visualization to Hierarchical and Multi-Output Labels. Academic Press.

Govindaraj, V., Thiyagarajan, A., Rajasekaran, P., Zhang, Y., & Krishnasamy, R. (2020). Automated unsupervised learn-
ing‐based clustering approach for effective anomaly detection in brain magnetic resonance imaging (MRI). IET Image
Processing, 14(14), 3516–3526. doi:10.1049/iet-ipr.2020.0597

Greco, S., Słowiński, R., & Szczęch, I. (2016). Measures of rule interestingness in various perspectives of confirmation.
Information Sciences, 346–347, 216–235. doi:10.1016/j.ins.2016.01.056

Gru¨tter, R. (2019). A framework for assisted proximity analysis in feature data. Journal of Geographical Systems, 21(3),
367–394. doi:10.100710109-019-00304-3

Guo, C., Zhang, M., & Chen, H. (2021). Suitability of low-field nuclear magnetic resonance (LF-NMR) combining with
back propagation artificial neural network (BP-ANN) to predict printability of polysaccharide hydrogels 3D printing.
International Journal of Food Science & Technology, 56(5), 2264–2272. doi:10.1111/ijfs.14844

Gupta, A., Sahayadhas, A., & Gupta, V. (2020). Proposed Techniques to Design Speed Efficient Data warehouse Archi-
tecture for Fastening Knowledge Discovery Process. 2020 IEEE Third International Conference on Artificial Intelligence
and Knowledge Engineering (AIKE).

Hannah, S., Deepa, A. J., Chooralil, V. S., BrillySangeetha, S., Yuvaraj, N., Arshath Raja, R., Suresh, C., Vignesh, R.,
YasirAbdullahR, Srihari, K., & Alene, A. (2022). Blockchain-Based Deep Learning to Process IoT Data Acquisition in
Cognitive Data. BioMed Research International, 2022, 1–7. doi:10.1155/2022/5038851 PMID:35187166

Haviluddin, I. M., Putra, G. M., Puspitasari, N., Setyadi, H. J., Dwiyanto, F.A., Wibawa, A. P., & Alfred, R. (2020). A
Performance Comparison ofEuclidean, Manhattan and Minkowski Distances in K-Means Clustering. 2020 6th Inter-
national Conference on Science in Information Technology (ICSITech), Science in Information Technology (ICSITech),
2020 6th International Conference On, 184–188.

Healy, K. (2018). Data visualization: a practical introduction. Princeton University Press.

Hernandez-Suarez, A., Sanchez-Perez, G., Toscano-Medina, K., Martinez-Hernandez, V., Sanchez, V., & Perez-Meana,
H. (2018). A Web Scraping Methodology for Bypassing Twitter API Restrictions. Academic Press.

Hill, C. (2015). Learning scientific programming with Python. Cambridge University Press. doi:10.1017/
CBO9781139871754

307
 Compilation of References

Hodijah, A., & Setijohatmo, U. T. (2021). Analysis of frequent itemset generation based on trie data structure in Apriori
algorithm. Telkomnika, 19(5), 1553–1564. doi:10.12928/telkomnika.v19i5.19273

Hong, G., & Nan, G. (2021). Research and Application of a Multidimensional Association Rules Mining Algorithm
Based on Hadoop. 2021 IEEE Intl Conf on Parallel & Distributed Processing with Applications, Big Data & Cloud
Computing, Sustainable Computing & Communications, Social Computing & Networking (ISPA/BDCloud/SocialCom/
SustainCom), Parallel & Distributed Processing with Applications, Big Data & Cloud Computing, Sustainable Com-
puting & Communications, Social Computing & Networking (ISPA/BDCloud/SocialCom/SustainCom), 2021 IEEE Intl
Conf on, ISPA-BDCLOUD-SOCIALCOM-SUSTAINCOM, 636–643.

Hongboonmee, N., & Trepanichkul, P. (2019). Comparison of Data Classification Efficiency to Analyze Risk Factors
that Affect the Occurrence of Hyperthyroidusing Data Mining Techniques. Mahanakorn University of Technology (MUT).

Hume, E., & West, A. (2020). Becoming a Data-Driven Decision Making Organization: Seven Challenges Facing Not-
for-Profits. The CPA Journal, 90(4), 32.

Hyrmet, M., & Arbana, K. (2021). The Data Mining Approach: A Case Study - Clustering Algorithms for After Sales
Service. 10th Mediterranean Conference on Embedded Computing (MECO), 1-6.

Inmon, W. H., & Lindstedt, D. (2015). Data architecture: a primer for the data scientist : big data, data warehouse and
data vault. Morgan Kaufmann/Elsevier.

Ironfrown. (2016). RapidMiner Stats (Part 5): Boxplots. Retrieved 15 December 2021 from https://www.youtube.com/
watch?v=IJ180B4ZiLo

Jamalludin, M. D. M,, Fajar Shidik, G., Zainul Fanani, A., Purwanto, & Al Zami, F. (2021). Implementation of Feature
Selection Using Gain Ratio Towards Improved Accuracy of Support Vector Machine (SVM) on Youtube Comment Clas-
sification. 2021 International Seminar on Application for Technology of Information and Communication (ISemantic),
Application for Technology of Information and Communication (ISemantic), 2021 International Seminar On, 28–31.

Jiang, S., Jiang, T., & Wang, L. (2020). Secure and Efficient Cloud Data Deduplication with Ownership Management.
IEEE Transactions on Services Computing, 13(6), 1152–1165.

Johnny, O., & Trovati, M. (2020). Big data inconsistencies and incompleteness: a literature review. International Journal
of Grid and Utility Computing, 5, 705. 10.1504/IJGUC.2020.110057

Joshi, P. (2017). Artificial intelligence with Python: a comprehensive guide to building intelligent apps for Python begin-
ners and developers. Packt Publishing Ltd.

Junaidi, A., Ferani Tanjung, N. A., Wijayanto, S., Lasama, J., & Iskandar, A. R. (2021). Overfitting Problem in Images
Classification for Egg Incubator Using Convolutional Neural Network. 2021 9th International Conference on Cyber and IT
Service Management (CITSM), Cyber and IT Service Management (CITSM), 2021 9th International Conference On, 1–7.

Jyothi, P. N., Lakshmi, D. R., & Rao, K. V. S. N. R. (2020). A Supervised Approach for Detection of Outliers in Health-
care Claims Data. Journal of Engineering Science & Technology Review, 13(1), 204–213. doi:10.25103/jestr.131.25

Kamukwamba, Y., & Chunxiao, L. (2021). A Novel Algorithm Using Content-Based Filtering Technology in Apache
Spark for Big Data Analysis. 2021 4th International Conference on Artificial Intelligence and Big Data (ICAIBD), Ar-
tificial Intelligence and Big Data (ICAIBD), 2021 4th International Conference On, 22–26.

Karthik, M. G., & Krishnan, M. B. M. (2021). Detecting Internet of Things Attacks Using Post Pruning Decision
Tree-Synthetic Minority Over Sampling Technique. International Journal of Intelligent Engineering & Systems, 14(4),
105–114. doi:10.22266/ijies2021.0831.10

308
Compilation of References

Keles, M. K. (2019). Breast Cancer Prediction and Detection Using Data Mining Classification Algorithms: A Compara-
tive Study. Tehnicki Vjesnik - Technical Gazette, 26(1), 149.

Khan, S. A., & Ali Rana, Z. (2019). Evaluating Performance of Software Defect Prediction Models Using Area Under
Precision-Recall Curve (AUC-PR). 2019 2nd International Conference on Advancements in Computational Sciences
(ICACS), Advancements in Computational Sciences (ICACS), 2019 2nd International Conference On, 1–6.

Khatter, H., Arif, S., Singh, U., Mathur, S., & Jain, S. (2021). Product Recommendation System for E-Commerce using
Collaborative Filtering and Textual Clustering. 2021 Third International Conference on Inventive Research in Computing
Applications (ICIRCA), Inventive Research in Computing Applications (ICIRCA), 2021 Third International Conference
On, 612–618.

Klinkenberg, R., & Hofmann, M. (2014). RapidMiner: data mining use cases and business analytics applications. CRC
Press.

Kotu, V., & Deshpande, B. (2014). Predictive analytics and data mining: concepts and practice with RapidMiner.
Elsevier Ltd.

Kumar, P., Ganapathy, G., & Kang, J.-J. (2021). A Hybrid Mod K-Means Clustering with Mod SVM Algorithm to
Enhance the Cancer Prediction. International Journal of Internet. Broadcasting and Communication, 13(2), 231–243.

Kusumawardhani, N. K., Nasrun, M., & Setianingsih, C. (2019). Web Recommended System Library Book Selection
Using Item Based Collaborative Filtering Method. 2019 IEEE International Conference on Engineering, Technology
and Education (TALE), Engineering, Technology and Education (TALE), 2019 IEEE International Conference On, 1–8.

Kyaw, K. S., & Limsiroratana, S. (2019). Case Study: Knowledge Discovery Process using Computation Intelligence
with Feature Selection Approach. 2019 17th International Conference on ICT and Knowledge Engineering (ICT&KE),
ICT and Knowledge.

Lakshmi Devi, R., & Samundeeswari, V., V. (2021). Detection and Automated Classification of Brain Tumor Types in
MRI Images using Convolutional Neural Network with Grid Search Optimization. 2021 Fifth International Conference
on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC), I-SMAC (IoT in Social, Mobile, Analytics and Cloud)
(I-SMAC), 2021 Fifth International Conference On, 1280–1284.

Lawal, M. M., Ibrahim, H., Sani, N. F. M., & Yaakob, R. (2020). Analyses of Indexing Techniques on Uncertain Data
With High Dimensionality. IEEE Access : Practical Innovations, Open Solutions, 8, 74101–74117. doi:10.1109/AC-
CESS.2020.2988487

LeCun, Y., Cortes, C., & Burges, J. C. C. (2021). The MNIST Database of Handwritten Digits. Retrieved from: http://
yann.lecun.com/exdb/mnist/

Li, J., Horiguchi, Y., & Sawaragi, T. (2020). Data Dimensionality Reduction by Introducing Structural Equation Modeling
to Machine Learning Problems. 2020 59th Annual Conference of the Society of Instrument and Control Engineers of Japan
(SICE), Instrument and Control Engineers of Japan (SICE), 2020 59th Annual Conference of the Society Of, 826–831.

Liao, S.-H., Widowati, R., & Chang, H.-Y. (2021). A data mining approach for developing online streaming recommen-
dations. Applied Artificial Intelligence, 35(15), 2204–2227. doi:10.1080/08839514.2021.1997211

Li, K., Li, J., Liu, S., Li, Z., Bo, J., & Liu, B. (2019). GA-iForest: An Efficient Isolated Forest Framework Based on
Genetic Algorithm for Numerical Data Outlier Detection. Transactions of Nanjing University of Aeronautics & Astro-
nautics, 36(6), 1026–1038.

309
 Compilation of References

Liu, L. (2021). Design of Human Resource Management Information System Based on Decision Tree Algorithm. 2021
Global Reliability and Prognostics and Health Management (PHM-Nanjing), Reliability and Prognostics and Health
Management (PHM-Nanjing), 2021 Global, 1–6.

Liu, Y.-Y., Li, L., Liu, Y.-S., Chan, P.-W., Zhang, W.-H., & Zhang, L. (2021). Estimation of precipitation induced by
tropical cyclones based on machine-learning-enhanced analogue identification of numerical prediction. Meteorological
Applications, 28(2). Advance online publication. doi:10.1002/met.1978

Lou, N. (2022). Analysis of the Intelligent Tourism Route Planning Scheme Based on the Cluster Analysis Algorithm.
Computational Intelligence and Neuroscience, 2022, 1–10. doi:10.1155/2022/3310676 PMID:35800698

Lourenco, J., & Varde, A. S. (2020). Item-Based Collaborative Filtering and Association Rules for a Baseline Recom-
mender in E-Commerce. 2020 IEEE International Conference on Big Data (Big Data), Big Data (Big Data), 2020 IEEE
International Conference On, 4636–4645.

Lubanovic, B. (2019). Introducing Python: modern computing in simple packages (2nd ed.). O’Reilly Media, Inc.

Mandhare, H. C., & Idate, S. R. (2017). A comparative study of cluster based outlier detection, distance based outlier
detection and density based outlier detection techniques. 2017 International Conference on Intelligent Computing and
Control Systems (ICICCS), 931-935. 10.1109/ICCONS.2017.8250601

Maneechaeye, P., & Ramjan, S. (2022). Components and Indicators Development with Guidelines to Encourage Safety
Behavior among Flight Crews in Thailand (Independent Study). Thammasat University.

Maslan, A., Mohamad, K. M. B., & Mohd Foozy, F. B. (2020). Feature selection for DDoS detection using classifica-
tion machine learning techniques. IAES International Journal of Artificial Intelligence, 9(1), 137–145. doi:10.11591/
ijai.v9.i1.pp137-145

Massaron, L., & Mueller, J. (2015). Python for data science for dummies. John Wiley and Sons, Inc.

Mat, T. M., Lajis, A., & Nasir, H. (2018). Text Data Preparation in RapidMiner for Short Free Text Answer in Assisted
Assessment. 2018 IEEE 5th International Conference on Smart Instrumentation, Measurement and Application (IC-
SIMA), 1-4.

Mesner, O. C., & Shalizi, C. R. (2021). Conditional Mutual Information Estimation for Mixed, Discrete and Continuous
Data. IEEE Transactions on Information Theory, 67(1), 464–484. doi:10.1109/TIT.2020.3024886

Mishra, A., & Vats, A. (2021). Supervised machine learning classification algorithms for detection of fracture location
in dissimilar friction stir welded joints. Frattura e Integrita Strutturale, 15(58), 242–253. doi:10.3221/IGF-ESIS.58.18

Mladenova, T. (2021). A Feature-Weighted Rule for the K-Nearest Neighbor. 2021 5th International Symposium on
Multidisciplinary Studies and Innovative Technologies (ISMSIT), Multidisciplinary Studies and Innovative Technologies
(ISMSIT), 2021 5th International Symposium On, 493–497.

Mnich, K., Polewko-Klim, A., Kitlas Golinska, A., Lesinski, W., & Rudnicki, W. R. (2020). Super Learning with Re-
peated Cross Validation. 2020 International Conference on Data Mining Workshops (ICDMW), Data Mining Workshops
(ICDMW), 2020 International Conference on, ICDMW, 629–635.

Mohan, L., Jain, S., Suyal, P., & Kumar, A. (2020). Data mining Classification Techniques for Intrusion Detection System.
2020 12th International Conference on Computational Intelligence and Communication Networks (CICN), 351-355.

Montesdeoca, B., Luengo, J., Maillo, J., García-Gil, D., García, S., & Herrera, F. (2019). A first approach on big data
missing values imputation. IoTBDS 2019 - Proceedings of the 4th International Conference on Internet of Things, Big
Data and Security, 315–323. 10.5220/0007738403150323

310
Compilation of References

Mueller, J. (2018). Beginning programming with Python (2nd ed.). John Wiley & Sons.

Mueller, J., & Massaron, L. (2019). Python for data science (2nd ed.). John Wiley & Sons, Inc.

Munlika, H. (2019). Machine Learning with Clustering Model. Retrieved June 17, 2021, from https://medium.com/
tniuniversity/%E0%B8%81%E0%B8%B2%E0%B8%A3%E0%B8%97%E0%B8%B3-machine-learning-%E0%B8%94
%E0%B9%89%E0%B8%A7%E0%B8%A2-clustering-model-2a3c392e7faa

Nan, X. (2022). Design and Implementation of a Personalized Tourism Recommendation System Based on the Data
Mining and Collaborative Filtering Algorithm. Computational Intelligence and Neuroscience, 1424097. PMID:36093493

Naparat, D. (2020). Data Mining with Rapidminer - Getting RapidMiner Educational License. Retrieved 15 December
2021 from https://www.youtube.com/watch?v=hBxjN1YYXXU

Nigam, H., Biswas, P., Raj, J. S., Iliyasu, A. M., Bestak, R., & Baig, Z. A. (2021). Web Scraping: From Tools to Related
Legislation and Implementation Using Python. Academic Press.

Nuttavut Thongjor. (2017). Understanding the data classification: k-Nearest Neighbours. Retrieved 9 August, 2021,
from https://www.babelcoder.com/blog/articles/k-nearest-neighbors

Padhy, S. K., Singh, A. K., & Vetrivelan, P. (2022). Item-Based Collaborative Filtering Blockchain for Secure Movie
Recommendation System. Lecture Notes in Electrical Engineering, 792, 937–948. doi:10.1007/978-981-16-4625-6_93

Pajankar, A. (2021). Practical Python data visualization: a fast track approach to learning data visualization with Python.
Apress. doi:10.1007/978-1-4842-6455-3

Paruechanon, P., & Sriurai, W. (2019). Applying association rule to risk analysis for student-dropout in Information
Technology Program. Journal of Science and Science Education, 1(2), 123-133.

Peungvicha, P., Tubtiang, A. & Sattayaaphitan, T. (2019). Clustering of Jewellery Purchasing Behaviour through Social
Network. Journal of Humanities and Social Sciences, Rajapruk University, 5, 212-224.

Phan, Q.-T., Wu, Y.-K., & Phan, Q.-D. (2021). An Overview of DataPreprocessing for Short-Term Wind Power Fore-
casting. 2021 7th International Conference on Applied System Innovation (ICASI), Applied System Innovation (ICASI),
2021 7th International Conference On, 121–125.

Punhani, R., Arora, V. P. S., Sabitha, S., & Kumar Shukla, V. (2021). Applicationof Clustering Algorithm for Effective
Customer Segmentation in E-Commerce. 2021 International Conference on Computational Intelligence and Knowledge
Economy (ICCIKE), Computational Intelligence and Knowledge Economy (ICCIKE), 2021 International Conference
On, 149–154.

Puspasari, B. D., Damayanti, L. L., Pramono, A., & Darmawan, A. K. (2021). Implementation K-Means Clustering
Method in Job Recommendation System. 2021 7th International Conference on Electrical, Electronics and Information
Engineering (ICEEIE), Electrical, Electronics and Information Engineering (ICEEIE), 2021 7th International Confer-
ence On, 1–6.

Puspitasari, D. I., Riza Kholdani, A. F., Dharmawati, A., Rosadi, M. E., & Mega Pradnya Dhuhita, W. (2021). Stroke Dis-
ease Analysis and Classification Using Decision Tree and Random Forest Methods. 2021 Sixth International Conference
on Informatics and Computing (ICIC), Informatics and Computing (ICIC), 2021 Sixth International Conference On, 1–4.

Rabiul Alam, M. G., Hussain, S., Islam Mim, M. M., & Islam, M. T. (2021). Telecom Customer Behavior Analysis
Using Naïve Bayes Classifier. 2021 IEEE 4th International Conference on Computer and Communication Engineering
Technology (CCET), Computer and Communication Engineering Technology (CCET), 2021 IEEE 4th International
Conference On, 308–312.

311
 Compilation of References

Rahman, B., Hendric Spits Warnars, H. L., Subirosa Sabarguna, B., & Budiharto, W. (2021). Heart Disease Classifica-
tion Model Using K-Nearest Neighbor Algorithm. 2021 Sixth International Conference on Informatics and Computing
(ICIC), Informatics and Computing (ICIC), 2021 Sixth International Conference On, 1–4.

Ramirez-Hernandez, J., Juarez-Sandoval, O.-U., Hernandez-Gonzalez, L., Hernandez-Ramirez, A., & Olivares-Dominguez,
R.-S. (2020). Voltage Control Based on a Back-Propagation Artificial Neural Network Algorithm. 2020 IEEE Interna-
tional Autumn Meeting on Power, Electronics and Computing (ROPEC), Power, Electronics and Computing (ROPEC),
2020 IEEE International Autumn Meeting On, 4, 1–6.

RapidMiner, Inc. (2020). How to Import Data into a Repository | RapidMiner Studio. Retrieved 15 December 2021 from
https://www.youtube.com/watch?v=DS-tYhgA5lA

Raschka, S., & Mirajalili, V. (2018). Python machine learning: machine learning and deep learning with Python, scikit-
learn, and TensorFlow (2nd ed.). Packt Publishing.

Rathaur, S., Kamath, N., & Ghanekar, U. (2020). Software Defect Density Prediction based on Multiple Linear Regression.
2020 Second International Conference on Inventive Research in Computing Applications (ICIRCA), Inventive Research
in Computing Applications (ICIRCA), 2020 Second International Conference On, 434–439.

Reges, S., Obourn, A., & Stepp, M. (2019). Building Python programs (1st ed.). Pearson.

Ren, X. (2021). Application of Apriori Association Rules Algorithm to Data Mining Technology to Mining E-commerce
Potential Customers. 2021 International Wireless Communications and Mobile Computing (IWCMC), Wireless Com-
munications and Mobile Computing (IWCMC), 2021 International, 1193–1196.

Rezaeijo, S. M., Hashemi, B., Mofid, B., Bakhshandeh, M., Mahdavi, A., & Hashemi, M. S. (2021). The feasibility of a
dose painting procedure to treat prostate cancer based on mpMR images and hierarchical clustering. Radiation Oncology
(London, England), 16(1), 1–16. doi:10.118613014-021-01906-2 PMID:34544468

Rezaei, M., Sanayei, A., Aghdaie, S. F. A., & Ansari, A. (2022). Improving the Omnichannel Customers’ Lifetime Value
Using Association Rules Data Mining: A Case Study of Agriculture Bank of Iran. Iranian Journal of Management
Studies, 15(1), 49–68.

Riswanto, E., Robi’in, B., & Suparyanto. (2019). Mobile Recommendation System for Culinary Tourism Destination using
KNN (K-nearest neighbor). Journal of Physics: Conference Series, 1201(1), 012039. doi:10.1088/1742-6596/1201/1/012039

Rogers, S., & Kalinova, E. (2021). Big Data-driven Decision-Making Processes, Real-Time Advanced Analytics, and
Cyber-Physical Production Networks in Industry 4.0-based Manufacturing Systems. Economics, Management, and
Financial Markets, 16(4), 84. doi:10.22381/emfm16420216

Romero, M. P., Parry, J., Prosser, A., Upton, P., Rees, E., Tearne, O., Arnold, M., Chang, Y.-M., Brunton, L. A., Stevens,
K., & Drewe, J. A. (2019). Decision tree machine learning applied to bovine tuberculosis risk factors to aid disease
control decision making. Preventive Veterinary Medicine, 175. PMID:31812850

Sabau Popa, C. D., Popa, D. N., Bogdan, V., & Simut, R. (2021). Composite Financial Performance Index Prediction
— A Neural Networks Approach. Journal of Business Economics and Management, 22(2), 277–296. doi:10.3846/
jbem.2021.14000

Salman, W. A. K., & Sadkhan, S. B. (2021). Proposed Association Rules Mining Algorithm for Sensors Data Streams.
2021 1st Babylon International Conference on Information Technology and Science (BICITS), Information Technology
and Science (BICITS), 2021 1st Babylon International Conference On, 76–81.

312
Compilation of References

Samasiri, P. (2020). Visual representations of different family structures using an interactive data exploration technique.
2020 1st International Conference on Big Data Analytics and Practices (IBDAP), Big Data Analytics and Practices
(IBDAP), 2020 1st International Conference On, 1–5.

Samsani, S. (2016). An RST based efficient preprocessing technique for handling inconsistent data. 2016 IEEE Interna-
tional Conference on Computational Intelligence and Computing Research (ICCIC), 1-8. 10.1109/ICCIC.2016.7919591

Shao, L., Mahajan, A., Schreck, T., & Lehmann, D. J. (2017). Interactive Regression Lens for Exploring Scatter Plots.
Computer Graphics Forum, 36(3), 157–166. doi:10.1111/cgf.13176

Sharma, M., Joshi, S., Sharma, S., Singh, A., & Gupta, R. (2021). Data Mining Classification Techniques to Assign
Individual Personality Type and Predict Job Profile. 2021 9th International Conference on Reliability, Infocom Technolo-
gies and Optimization (Trends and Future Directions) (ICRITO), 1-5.

Shrestha, R. B., Razavi, M., & Prasad, P. W. (2020). An Unsupervised Machine Learning Technique for Recommenda-
tion Systems. 2020 5th International Conference on Innovative Technologies in Intelligent Systems and Industrial Ap-
plications (CITISIA), Innovative Technologies in Intelligent Systems and Industrial Applications (CITISIA), 2020 5th
International Conference On, 1–9.

Sneeringer, L. (2016). Professional Python. WROX/John Wiley & Sons.

Snineh, S. M., Amrani, N. E. A., Youssfi, M., Bouattane, O., & Daaif, A. (2021). Detection of traffic anomaly in high-
ways by using recurrent neural network. 2021 Fifth International Conference On Intelligent Computing in Data Sciences
(ICDS), Intelligent Computing in Data Sciences (ICDS), 2021 Fifth International Conference On, 1–6.

Soft, N. (2020). RapidMiner Tutorial For Beginners | Download & Install | Introduction to RapidMiner. Retrieved 15
December 2021 from https://www.youtube.com/watch?v=im11NBJfjhM

Song, H., Fu, Y., Saket, B., & Stasko, J. (2021). Understanding the Effects of Visualizing Missing Values on Visual Data
Exploration. 2021 IEEE Visualization Conference (VIS), Visualization Conference (VIS), 2021 IEEE, VIS, 161–165.

Song, J., Yu, Q., & Bao, R. (2019). The Detection Algorithms for Similar Duplicate Data. 2019 6th International Con-
ference on Systems and Informatics (ICSAI), Systems and Informatics (ICSAI), 2019 6th International Conference On,
1534–1542.

Sonu, S. B., & Suyampulingam, A. (2021). Linear Regression Based Air Quality Data Analysis and Prediction using
Python. 2021 IEEE Madras Section Conference (MASCON), Madras Section Conference (MASCON), 1–7.

SPSS Inc. and International Consortium Publish CRISP-DM 1.0; Free Document Offers Step-By-Step Guidance for
Data Mining Projects. (2000). Business Wire.

Sriwiboon, N. (2020). Improvement the Performance of the Chest X-ray Image Classification with Convolutional Neural
Network Model by Using Image Augmentations Technique for COVID-19 Diagnosis. The Journal of King Mongkut’s
University of Technology North Bangkok, 31(1).

Stewart, J. (2017). Python for scientists (2nd ed.). Cambridge University Press. doi:10.1017/9781108120241

Sunkpho, J. & Ramjan, S. (2020). Predicting Condominium Price in Bangkok Using Web Mining Techniques. Sri-
nakharinwirot Research and Development (Journal of Humanities and Social Sciences), 12(24).

Sunkpho, J., & Ramjan, S. (2020). Predicting Condominium Price in Bangkok using Web Mining Technique. Srinakharin-
wirot Research and Development (Journal of Humanities and Social Sciences), 12(24).

313
 Compilation of References

Sunkpho, J., & Hofmann, M. (2019). Analyzing Customer Satisfaction of a Mobile Application using Data Mining
Techniques. Research Administration Division, Thammasat University.

Sunkpho, S., & Ramjan, S. (2020). Introduction to Python Programming for Data Science. Thammasat Press.

Susanti, S. P., & Azizah, F. N. (2017). Imputation of missing value using dynamic Bayesian network for multivariate time series
data. 2017 International Conference on Data and Software Engineering (ICoDSE), 1-5. 10.1109/ICODSE.2017.8285864

Tai, Y., Sun, Z., & Yao, Z. (2021). Content-Based Recommendation Using Machine Learning. 2021 IEEE 31st Interna-
tional Workshop on Machine Learning for Signal Processing (MLSP), Machine Learning for Signal Processing (MLSP),
2021 IEEE 31st International Workshop On, 1–4.

Tananatong, T., & Ramjan, S. (2022). Data Visualization Design and Development for Data Science. Thammasat Press.

Tanantong, T. (2021). Data Mining. Retrieved from: https://www.skilllane.com/courses/tuxsa-Data-Mining-Algorithms/

chapter

Tanantong, T., & Ramjan, S. (2021, April-June). An Association Rule Mining Approach to Discover Demand and Sup-
ply Patterns Based on Thai Social Media Data. International Journal of Knowledge and Systems Science, 12(2), 1–16.
doi:10.4018/IJKSS.2021040101

Tanantong, T., & Ramjan, S. (2022). Data Visualization for Data Science (1st ed.). Thammasat Press.

Telikani, A., Shahbahrami, A., & Gandomi, A. H. (2021). High-performance implementation of evolutionary privacy-
preserving algorithm for big data using GPU platform. Information Sciences, 579, 251–265. doi:10.1016/j.ins.2021.08.006

Thammasisir, S., & Ramjan, S. (2021). Corrosion under Insulation severity Prediction for Carbon Steel in Marine
Environment using Data Science Methodology (Independent Study). Thammasat University.

Thangakumar, J., & Kommina, S. B. (2020). Ant colony optimization based feature subset selection with logistic regres-
sion classification model for education data mining. International Journal of Advanced Science and Technology, 29(3),
5821–5834.

Thankachan, T., Prakash, K. S., & Jothi, S. (2021). Artificial neural network modeling to evaluate and predict the me-
chanical strength of duplex stainless steel during casting. Sadhana, 46(4), 1–12. doi:10.100712046-021-01742-w

Torrens, M., & Tabakovic, A. (2022). A Banking Platform to Leverage Data Driven Marketing with Machine Learning.
Entropy (Basel, Switzerland), 24(3), 347. doi:10.3390/e24030347 PMID:35327858

Toumi, A., Gribaa, N., & Ben Abdessalem Karaa, W. (2021). Mining biomedical texts based on statistical method and
association rules. 2021 International Conference of Women in Data Science at Taif University (WiDSTaif), Women in
Data Science at Taif University (WiDSTaif), 2021 International Conference Of, 1–4.

Tuition, D. (2021). RapidMiner Tutorial, Learn Data Cleaning, Data Visualization & Data Analysis with Rapidminer.
Retrieved 15 December 2021 from https://www.youtube.com/watch?v=r27eUHJtIFs

Van der Ark, L. A., Croon, M. A., & Sijtsma, K. (2005). New developments in categorial data analysis for the social
and behavioral sciences. Lawrence Erlbaum Associates Publishers.

Vasudevan, S. (2021). Rapid Miner Installation - Step by Step explanation. Retrieved 15 December 2021 from https://
www.youtube.com/watch?v=Wp42fV4q66Y

Vichi, M., Ritter, G., & Giusti, A. (2013). Classification and data mining. Springer Verlag.

314
Compilation of References

Wafa, H. A., Aminuddin, R., Ibrahim, S., Mangshor, N. N. A., & Wahab, N. I. F. A. (2021). A Data Visualization Frame-
work during Pandemic using the Density-Based Spatial Clustering with Noise (DBSCAN) Machine Learning Model.
2021 IEEE 11th International Conference on System Engineering and Technology (ICSET), System Engineering and
Technology (ICSET), 2021 IEEE 11th International Conference On, 1–6.

Wang, G., & Chen, Y. (2022). Enabling Legal Risk Management Model for International Corporation with Deep Learn-
ing and Self Data Mining. Computational Intelligence and Neuroscience, 2022, 6385404. doi:10.1155/2022/6385404
PMID:35432517

Wang, J., Sato, K., Guo, S., Chen, W., & Wu, J. (2019). Big Data Processing With Minimal Delay and Guaranteed
Data Resolution in Disaster Areas. IEEE Transactions on Vehicular Technology, 68(4), 3833–3842. doi:10.1109/
TVT.2018.2889094

Wayan Priscila Yuni Praditya, N., Erna Permanasari, A., & Hidayah, I. (2021). Designing a tourism recommendation
system using a hybrid method (Collaborative Filtering and Content-Based Filtering). 2021 IEEE International Confer-
ence on Communication, Networks and Satellite (COMNETSAT), Communication, Networks and Satellite (Comnetsat),
2021 IEEE International Conference On, 298–305.

Wijaya, Y. A., Kurniady, D. A., Setyanto, E., Tarihoran, W. S., Rusmana, D., & Rahim, R. (2021). Davies Bouldin
Index Algorithm for Optimizing Clustering Case Studies Mapping School Facilities. TEM Journal, 10(3), 1099–1103.
doi:10.18421/TEM103-13

Xu, B., Chen, C., & Yang, J.-H. (2022). Application of Cluster Analysis Technology in Visualization Research of Movie
Review Data. Computational Intelligence and Neuroscience, 2022, 1–11. doi:10.1155/2022/7756896 PMID:35880060

Yadalam, T. V., Gowda, V. M., Kumar, V. S., Girish, D., & M., N. (2020). Career Recommendation Systems using Content
based Filtering. 2020 5th International Conference on Communication and Electronics Systems (ICCES), Communica-
tion and Electronics Systems (ICCES), 2020 5th International Conference On, 660–665.

Yager, R. R. (2020). Ordinal scale based uncertainty models for AI. Information Fusion, 64, 92–98. doi:10.1016/j.
inffus.2020.06.010

Yang, F.-J. (2019). An Extended Idea about Decision Trees. 2019 International Conference on Computational Science
and Computational Intelligence (CSCI), 349-354. 10.1109/CSCI49370.2019.00068

Yang, H., Zheng, Z., & Sun, C. (2022). E-Commerce Marketing Optimization of Agricultural Products Based on Deep
Learning and Data Mining. Computational Intelligence and Neuroscience, 2022, 2022. doi:10.1155/2022/6564014
PMID:35634060

Yingzhuo, X., & Xuewen, W. (2021). Research on Community Consumer Behavior Based on Association Rules Analysis.
2021 6th International Conference on Intelligent Computing and Signal Processing (ICSP), Intelligent Computing and
Signal Processing (ICSP), 2021 6th International Conference On, 1213–1216.

Zadeh, A. H., Zolbanin, H. M., & Sharda, R. (2021). Incorporating Big Data Tools for Social Media Analytics in a Busi-
ness Analytics Course. Journal of Information Systems Education, 32(3), 176–198.

Zagatti, F. R., Silva, L. C., Dos Santos Silva, L. N., Sette, B. S., Caseli, H. de M., Lucredio, D., & Silva, D. F. (2021).
MetaPrep: Data preparation pipelines recommendation via meta-learning. 2021 20th IEEE International Conference on
Machine Learning and Applications (ICMLA), Machine Learning and Applications (ICMLA), 2021 20th IEEE Interna-
tional Conference on, ICMLA, 1197–1202.

Zelle, J. M. (2010). Python programming: an introduction to computer science (2nd ed.). Beedle & Associates.

315
 Compilation of References

Zhang, H., Zhang, L., & Jiang, Y. (2019). Overfitting and Underfitting Analysis for Deep Learning Based End-to-end
Communication Systems. 2019 11th International Conference on Wireless Communications and Signal Processing
(WCSP), Wireless Communications and Signal Processing (WCSP), 2019 11th International Conference On, 1–6.

Zhang, C., & Zhang, S. (2002). Association Rule Mining. In Models and Algorithms. Springer Berlin Heidelberg.
doi:10.1007/3-540-46027-6

Zhou, Y., Yan, S., Wang, C., Zheng, K., & Zhu, L. (2021). Position Monitoring System Based on Hierarchical Cluster-
ing. 2021 26th IEEE Asia-Pacific Conference on Communications (APCC), Communications (APCC), 2021 26th IEEE
Asia-Pacific Conference On, 111–114.

Zhu, H., Samtani, S., Brown, R. A., & Chen, H. (2021). A Deep Learning Approach for Recognizing Activity of Daily
Living (Adl) for Senior Care: Exploiting Interaction Dependency and Temporal Patterns. Management Information
Systems Quarterly, 45(2), 859–895. doi:10.25300/MISQ/2021/15574

Zuo, L., & Guo, J. (2019). Customer Classification of Discrete Data Concerning Customer Assets Based on Data Mining.
2019 International Conference on Intelligent Transportation, Big Data & Smart City (ICITBS). 10.1109/ICITBS.2019.00093

316
 317

About the Authors

Sarawut Ramjan is a highly accomplished academician with a Ph.D. in Computer and Engineering
Management, which he earned from Assumption University in 2013. In 2017, he joined the College of
Innovation at Thammasat University as an Associate Professor, where he currently conducts cutting-
edge research at the Thammasat University AI Center. Dr. Ramjan is a renowned expert in data science,
with a strong focus on teaching and research in this field. His remarkable contributions to the academic
world have made him a prominent figure in the academic community.

Jirapon Sunkpho is a distinguished scholar with a passion for academic excellence. He holds a Ph.D.
in Civil and Environmental Engineering (Computer-Aided Engineering) from the esteemed Carnegie
Mellon University, earned in 2001, as well as an MSc in Applied Data Science and Analytics from
Technological University Dublin, which he completed in 2021. Dr. Sunkpho is a visionary leader and
currently serves as the Vice Rector in Information Technology at Thammasat University, where he plays
a key role in shaping the institution’s technological direction. He is also a Board of Director at Airport of
Thailand, where his expertise is valued. As an accomplished Associate Professor, Dr. Sunkpho is widely
recognized for his teaching excellence and his research in data science. His contributions to this field
have been profound, and he remains committed to advancing knowledge in this area. With his unique
combination of academic prowess and leadership abilities, Dr. Sunkpho is a true asset to the academic
and professional communities.


318

Index

A 76-83, 87, 92, 95, 97-98, 101-106, 108, 133-134,

136, 138-139, 147, 153, 155-158, 160, 165, 173-
Apriori Algorithm 160, 162-164, 166, 175, 198 175, 178, 191, 193-195, 197, 199-207, 210, 235,
artificial neural network 105, 108-114, 117, 119, 129, 237-238, 253, 272, 274, 276-277, 281, 287, 291
136-138 data mining techniques 1-8, 10, 15-16, 18, 20, 25, 27,
association rules 9, 160-162, 164-168, 171-172, 174- 29, 32, 35-36, 39-40, 42, 45-47, 50-51, 55, 63,
176, 191 65, 68, 76-81, 87, 101-105, 107-108, 133-134,
136, 147, 153, 155-157, 160, 174, 178, 193-195,
B 197, 199, 201, 203-206, 210, 237-238, 253, 272,
274, 276-277, 287
Big Data 1, 3, 11-12, 15, 28-30, 32-34, 45-49, 62, data points 141-148, 150
80-82, 103, 107, 129, 134, 136, 138, 145, 152, Data scientists 1-5, 7-12, 26-27, 30, 32, 35-43, 47, 50,
157-158, 160, 175, 191, 202, 235, 291 55-56, 58, 62-63, 65-70, 72, 75-81, 83, 85-95, 97-
104, 107-108, 110-113, 115-125, 127, 129-130,
C 133-136, 139-153, 155, 157, 160-162, 164-168,
174, 177-181, 183-186, 195, 199, 203, 205-207,
Cloud Technology 1, 3 209-233, 235, 237-248, 250-270, 272-277, 279-
Confidence Thresholds 161-165, 198 280, 282-291
Convolutional Operation 117-119, 123, 125 data storage 4, 13, 274-275
Cross Validation 84-85, 87, 96, 98-100, 102, 104, decision making 1, 3, 13, 34, 46, 48, 83, 103, 143,
106, 116 165, 168, 174
Cross-industry Standard 4, 6 Decision Tree 7-8, 20, 28, 34, 72, 75-77, 79, 81, 83,
customer satisfaction 193-195, 204 89-90, 92, 94, 97-106, 134, 147, 157, 195, 197,
customer segmentation 139, 155, 157, 159 200, 277, 279-280, 282, 292
Deep learning 2-3, 8, 11, 18, 106-108, 117, 121-123,
D 125, 127-138, 191, 197, 236, 276
digital device 173, 206, 215-216, 221
data analysis 1-7, 10, 12-18, 20, 22-32, 35-40, 42-46, Digital Transformation 20, 42, 155-156, 216, 218, 229
49-51, 55-56, 62-64, 67, 72, 76, 79-81, 84, 87, Domain Experts 35, 50, 62, 78, 139, 160, 201
89, 92, 101-102, 104, 108-109, 114, 117, 133,
136, 140, 152-153, 155-158, 162, 164, 166-169,
174-175, 185-186, 190-191, 193-195, 197, 199-
E
204, 206, 211, 237, 257, 276-277, 279-283, 285, Elective Courses 15-18, 20
287-288, 291-292
Data classification 50, 65, 72, 83, 85-87, 89-90, 92, F
94, 97-98, 100-101, 104-106, 117-118, 122, 136,
147, 200, 277-280 Feature Extraction 105, 117, 119, 287-288
data mining 1-8, 10-12, 15-16, 18, 20, 25, 27, 29, 32- filter matrix 117-119, 124
36, 39-40, 42, 45-47, 50-51, 55-56, 63-66, 68,



Index

K R
K-Means clustering 140, 142, 148-150, 153, 157-159 RapidMiner software 10, 32, 39, 47, 50-55, 57, 63,
K-Nearest Neighbor (k-NN) 33, 83, 87, 106, 191-192 65-66, 68, 72, 76, 81, 92, 94, 98, 100, 104, 112-
Knowledge Discovery 3-4, 6, 33 113, 115, 121-122, 129, 131, 136, 148-149, 157,
166-167, 174, 184-187, 194-195, 197, 203-204
L recommendation system 159, 175, 177-178, 183-186,
188, 190-192, 202-204
learning techniques 2, 33, 107, 121, 123, 129, 133, 136 Regression Analysis 15-16, 18, 62-63, 276-278, 281-
LinearRegression 277, 281-282, 285, 287 283, 285-288, 291

M S
machine learning 1, 3, 5, 8, 11-12, 15, 23, 25, 27-28, Safe Driving 193-194, 201
32-34, 44, 48-49, 72, 75-76, 81-83, 85, 87-89, Scatter Matrix 280
93, 97, 101, 103-105, 108, 113, 116, 121-122, Scatter Plot 38, 41, 140, 143-145, 199, 272-273, 277,
125, 127, 129-130, 135-136, 139, 159, 184-185, 282, 284-285, 287
188, 191-192, 195-196, 199-201, 236, 276-277, Scikit-Learn library 27, 276-277, 288
279-281, 285 Social Networks 1, 4-5, 10, 44, 155, 193
Manhattan Distance 142, 158 Statistical Learning 12, 15-16, 18
mining techniques 1-10, 15-16, 18, 20, 25, 27, 29, 32, Supervised Learning 7, 10, 32, 65, 72, 83, 107, 276-277
34-36, 38-40, 42, 45-47, 50-51, 55, 63, 65, 68,
76-81, 87, 101-105, 107-108, 133-134, 136, 147, T
153, 155-157, 160, 174, 178, 193-195, 197, 199,
201, 203-206, 210, 237-238, 246, 253, 272, 274, Taxable Income 90
276-277, 287 teaching machines 7, 83, 87, 122, 184, 186, 197
multiple regression 201, 281-283, 285
U
N
Unsupervised Learning 8, 10, 32, 65, 139, 158, 160,
Nearest-Neighbor Collaborative 177, 179-180 193, 237
neural network 20, 27, 30-31, 102, 105, 107-114, 116- Unsupervised-Learning data 139
117, 119-120, 129, 133, 136-138, 292 User-Based Nearest-Neighbor 179-180
Non-Linear Function 108, 118, 121
V
P
visual information 5, 10, 67-68, 150
polynomial regression 5, 281, 285-288, 291-292
Predicting Condominium Price 34, 193, 195, 204
programming skills 10, 23, 50, 63, 65
purchasing behavior 134, 140, 142-143, 160, 177-178

319