Content uploaded by Viswanatha v.
Author content
All content in this area was uploaded by Viswanatha v. on Aug 07, 2023
Content may be subject to copyright.
International Journal of Engineering and Management Research Peer Reviewed & Refereed Journal
e-ISSN: 2250-0758 | p-ISSN: 2394-6962 Volume-13, Issue-4 (August 2023)
https://ijemr.vandanapublications.com https://doi.org/10.31033/ijemr.13.4.6
45
This work is licensed under Creative Commons Attribution 4.0 International License.
Online Fraud Detection Using Machine Learning Approach
Viswanatha V1, Ramachandra A.C2, Deeksha V3 and Ranjitha R4
1Assistant Professor, Department of Electronics and Communication Engineering, Nitte Meenakshi Institute of Technology,
Bangalore, INDIA
2Professor, Department of Electronics and Communication Engineering, Nitte Meenakshi Institute of Technology, Bangalore,
INDIA
3Student, Department of Electronics and Communication Engineering, Nitte Meenakshi Institute of Technology, Bangalore,
INDIA
4Student, Department of Electronics and Communication Engineering, Nitte Meenakshi Institute of Technology, Bangalore,
INDIA
1Corresponding Author: viswas779@gmail.com
Received: 04-07-2023 Revised: 20-07-2023 Accepted: 04-08-2023
ABSTRACT
Online extortion discovery has ended up a
tremendous issue in today’s advanced age and poses a danger
to individuals, businesses, and budgetary teachers all over the
world. The increment in extortion illustrates the require for
compelling extortion discovery, especially within the setting of
anti-money laundering (AML) endeavors. This extent is
planned to create a machine learning based arrangement
utilizing Python to distinguish and avoid online extortion in
genuine time.
The proposed framework employment chronicled
exchange information, combining different components such
as client behavior, exchanges, and budgetary information.
First, the information control prepare is utilized to clean the
information and change over it into organize reasonable for
the preparing show. At that point, different machine learning
calculations such as calculated relapse, choice trees, irregular
timberlands or angle boosting are used to build predictive
algorithms that can spot fraud. The extended concludes with
the usage of the created show in a genuine world online
exchange environment, permitting for genuine time extortion
location and avoidance. The system’s adequacy is persistently
checked and assessed, and essential overhauls and
advancements are made to adjust to advancing extortion
designs and procedures. By and large, this extends points to
supply a strong and proficient arrangement utilizing Python
and machine learning strategies to combat online extortion.
By precisely recognizing false exchanges in genuine time, this
framework can altogether contribute to fortifying AML
endeavors and ensuring people and organizations from
money related misfortunes and reputational harm related
with online extortion.
Keywords-- Unique Information Mining, Online Fraud
Detection, Machine Learning, Decision Tree Algorithm
Ⅰ. INTRODUCTION
Online extortion has become a predominant issue
in today's advanced age. With the expanding dependence
on innovation and the web for different exchanges,
hoodlums have found modern and advanced ways to
misdirect and dupe clueless people and businesses. The
result is required for successful extortion discovery
frameworks has ended up more vital than ever some time
recently. One such structure has risen as effective
apparatus in combating online extortion is online extortion
detection[1]-[2]
Online extortion discovery alludes to the utilize of
progressed calculations and machine learning procedures
to distinguish and anticipate false exercises happening over
the web. It includes analyzing huge volumes of
information in real-time to identify designs,
inconsistencies, and suspicious behavior that will
demonstrate false action. By leveraging fake insights and
information analytics, online extortion location
frameworks can rapidly distinguish potential fraudsters
and take fitting activities to relieve the dangers related with
online extortion [3]-[4].
The presentation of online extortion locations has
revolutionized the way organizations approach extortion
avoidance. Conventional strategies of extortion discovery,
such as manual audits and rule-based frameworks, were
regularly time-consuming, incapable, and inclined to
human mistake. These strategies depended intensely on
human instinct and were incapable to keep up with the
quickly advancing strategies utilized by fraudsters [[4]-[5]
In differentiate, online extortion discovery frameworks are
competent of handling endless sums of information in real-
time, permitting for prompt distinguishing proof and
reaction to false exercises. These frameworks can analyze
different information focuses at the same time, counting
International Journal of Engineering and Management Research Peer Reviewed & Refereed Journal
e-ISSN: 2250-0758 | p-ISSN: 2394-6962 Volume-13, Issue-4 (August 2023)
https://ijemr.vandanapublications.com https://doi.org/10.31033/ijemr.13.4.6
46
This work is licensed under Creative Commons Attribution 4.0 International License.
client behavior, exchange history, gadget data, and
relocation information, to construct a comprehensive
profile of each client. By comparing this profile against
known designs of false behavior, these frameworks can
precisely distinguish and hail suspicious exercises [6]-[7].
Because of the emergence of cutting-edge innovation, the
mode of payment of a person has changed altogether. The
utilize of Online Installment mode such as Online
Managing an account, Charge Card, Credit Card etc. has
ended up well known and is getting to be important in day-
to-day exercises since it permits bank customers to buy
products and administrations from the shopping websites.
Figure.1: Online fraud detection
Figure 1 shows the process involved in online
fraud detection and some techniques used in online fraud
detection include machine learning, behavioral analysis,
and data mining.
Extortion bargains with cases that happen due to
criminal reasons which are troublesome to recognize.
Extortion can be basically partitioned into two types:
Most of the offline extortion
episodes happen due to the taking of purse/wallet that
contains critical documents. Archives such as Driving
Permit, ID card etc. contain pivotal data such as title, date
of birth, exchange slips etc.
Online extortion happens when
fraudster present their site as a veritable site in arrange to
obtain pivotal individual information of a client and
perform illegal exchanges on such client account.
Online extortion discovery employing a decision
tree classifier includes employing a choice tree calculation
to classify whether a transaction or movement is false or
not. The method ordinarily includes the taking after steps:
1. Information Collection: Collect significant information
related to online exchanges or exercises, counting
highlights such as exchange sum, IP address, area, gadget
data, client behavior designs, etc.
2. Information Preprocessing: Clean and preprocess the
collected information by dealing with lost values, outliers,
and changing categorical factors into numerical
representations.
3. Include Determination: Select the foremost pertinent
highlights that are likely to contribute to extortion location.
This step makes a difference in decreasing the
dimensionality of the information and making strides the
model's performance.
4. Part the Information: Part the preprocessed
information into preparing and testing sets. The training set
is utilized to prepare the choice tree classifier, whereas the
testing set is utilized to assess its performance.
5. Choice Tree Training: Train the choice tree classifier
on the preparing information. The calculation will learn
designs and rules from the information to form
expectations approximately extortion.
6. Show Assessment: Assess the execution of the prepared
choice tree classifier utilizing different assessment
measurements such as precision, accuracy, review, and F1-
score. This step makes a difference in evaluating the
adequacy of the demonstration in identifying online fraud.
7. Hyper parameter Tuning: Optimize the choice tree
classifier by tuning its hyper parameters to progress its
execution encouragement. Common hyper parameters
incorporate greatest profundity, least tests for a part, and
measure for splitting.
8. Foreseeing Extortion: Utilize the prepared choice tree
classifier to foresee whether a modern exchange or action
is false or not based on its features.
9. Checking and Overhauling: Ceaselessly screen the
model's execution and overhaul it with modern information
to adjust to changing extortion designs and move forward
exactness.
Figure.2: Flowchart of online fraud detection
Figure 2 represents the flowchart of online fraud
detection from information collection and preprocessing to
checking and overhauling.
Strategies to Take Individual Information
There are different strategies or procedures that
cyber hoodlums used to commit the crime:
StarT Data Collection Data
Preprocessing Feature
Selection
Splitting the
Data Decision Tree
Training Model
Evaluation Hyperparameter
Tuning
Predicting Fraud Monitoring and
Updating End
International Journal of Engineering and Management Research Peer Reviewed & Refereed Journal
e-ISSN: 2250-0758 | p-ISSN: 2394-6962 Volume-13, Issue-4 (August 2023)
https://ijemr.vandanapublications.com https://doi.org/10.31033/ijemr.13.4.6
47
This work is licensed under Creative Commons Attribution 4.0 International License.
Hacking: Programmer could be a individual who
looks for and misuses weakness in computer
framework. Aggressors break into industry or
individual databases.
Phishing: Phishing could be a false endeavor,
not many through e-mail, to take your individual
information.
Spoofing: The word "parody" implies hoax, trap,
or deceive. Spoofing alludes to deceiving or
deluding the computer clients. Usually ordinarily
done by covering up one's identity or faking the
personality of another client on the Internet.
Spyware: The computer client unconsciously
downloads software from the Web that contains
spyware. Spyware collects individual data from
your computer and transmits it to fraudsters or
attackers.
Bear Surfing: An assailant observes a bank client
from a adjacent area as the client punches in his
personal data. In the event that the client is giving
his personal data over the phone (e.g., to a inn or
car rental company), the assailant may tune in to
the conversation so as to get individual data of
bank customer
Dumpster Jumping: An aggressor goes through a
customer’s trash cans or waste containers to get
individual information of bank client such as bank
articulation, payment receipt etc.
Ⅱ. LITERATURE REVIEW
The outbreak of COVID-19 burgeons newborn
services on online platforms and simultaneously buoys
multifarious online fraud activities. Due to the rapid
technological and commercial innovation that opens an
ever-expanding set of products, insufficient labeling data
renders existing supervised or semi-supervised fraud
detection models ineffective in these emerging services.
However, the ever-accumulated user behavioral data on
online platforms might be helpful in improving
Investigating fraud in children's services. For this purpose,
in this article, we first present the user behavior, which
consists of orderly arranged actions, from the large-scale
unlabeled data sources for online fraud detection. Recent
studies illustrate that accurate extraction of user intentions
(formed by consecutive actions) in behavioral sequences
can propel improvements in the performance of online
fraud detection. analyze the nature of online fraud; We are
developing a model called UB-PTM that investigates
fraudulent knowledge. Actions on three institutional tasks
with varying degrees of detail, i.e., actions, The level of
intent and consistency of large untagged data. Extensive
experimentation with three downstream tasks to detect
online fraud at the transaction and user level has resulted
in our UB-PTM It can outperform modern designs for
certain tasks [8]-[9]
Increased use of computer innovations and
continued business development Exchanging money
through e-commerce systems such as credit card systems,
multimedia transmission systems, health protection
systems, etc. Unfortunately, these systems are used by both
real customers and fraudsters. Scammers also use different
approaches to hack e-commerce platforms. Fraud
Prevention System (FPS) cannot guarantee: Legitimate
security of e-commerce systems. In any case, the
interaction between FPS and FDS can help secure e-
commerce systems. All things considered, there are issues
and challenges that prevent the execution of FDSs, such as
concept float, underpins genuine time location, skewed
conveyance, expansive sum of information etc. This study
paper points to supply a precise and comprehensive outline
of these issues and challenges that deter the execution of
FDSs. We have chosen five electronic commerce
frameworks, which are credit card, media transmission,
healthcare protections, car protections and online sell off.
The predominant extortion sorts in those E-commerce
frameworks are presented closely. Encourage, state-of-the-
art FDSs approaches in chosen E-commerce frameworks
are efficiently presented. At that point a brief talk on
potential investigate patterns within the close future and
conclusion are displayed.[10]
Progressively, customers depend on social data
channels, such as user-posted online audits, to form buy
choices. These audits are accepted to be fair-minded
reflections of other consumers' encounters with the items
or administrations. Whereas broadly accepted, the writing
has not tried the presence or non-existence of survey
control. By utilizing information from Amazon and Barnes
& Respectable, our ponder examines if merchants,
distributors, and scholars reliably control online customer
surveys. We record the presence of online survey control
and appear that the control technique of firms appears to
be a monotonically diminishing work of the product's
genuine quality or the cruel buyer rating of that item.
Control in this way diminishes the instructive substance of
online surveys. Shoppers moreover get it the presence of
control but can as it were mostly adjusted it based on their
desires of the general level of control. Hence, merchants
can manipulate online analysts to alter the ultimate comes
about. It moreover appears that both price and surveys
serve as pointers of quality within the early stages of a
product's launch on the Amazon commercial center. So, at
this arrangement, higher costs increment deals instead of
diminishing them. In the afterward stages, costs play a
ordinary part. In other words, higher costs lead to lower
deals. At long last, Barnes & Noble's normal control level
is higher than Amazon's [11]-[12].
International Journal of Engineering and Management Research Peer Reviewed & Refereed Journal
e-ISSN: 2250-0758 | p-ISSN: 2394-6962 Volume-13, Issue-4 (August 2023)
https://ijemr.vandanapublications.com https://doi.org/10.31033/ijemr.13.4.6
48
This work is licensed under Creative Commons Attribution 4.0 International License.
Nowadays illicit exercises with respect to online
money related exchanges have ended up progressively
complex and borderless, coming about in colossal
budgetary misfortunes for both sides, clients, and
organizations. Numerous strategies have been proposed to
extortion anticipation and discovery within the online
environment. Be that as it may, all these strategies other
than having the same objective of recognizing and
combating false online exchanges, they come with their
possess characteristics, focal points and drawbacks. In this
setting, this paper reviews the existing inquiry about
drained extortion discovery with the point of recognizing
calculations utilized and analyzing each of these
calculations based on certain criteria. To analyze the
investigation ponders within the field of extortion
discovery, the efficient quantitative writing audit strategy
was connected. A progressive typology is built based on
the machine learning calculations most frequently
mentioned in logical papers and their characteristics. did.
Therefore, our article highlights the most excellent
strategies for discovery in a other way. It anticipates
extortion by combining three determination criteria:
financial soundness, scope, and taken a toll [13].
Utilizing remote versatile terminals has gotten to
be the standard of Web exchanges, which can confirm the
character of clients by passwords, fingerprints, sounds, and
pictures. In any case, once this personality information is
stolen, conventional data security strategies will not dodge
online exchange extortion. The existing convolutional
neural organize show for extortion location should create
numerous subordinate highlights. This paper proposes a
extortion discovery demonstrate based on the
convolutional neural arrange within the field of online
exchanges, which develops an input highlight sequencing
layer that actualizes the reorganization of crude exchange
highlights to create diverse convolutional designs. Its
centrality is that distinctive combinations entering the
convolution bit will create diverse subordinate highlights.
The advantage of this model is that it takes low-
dimensional and non-derivative online value-based
information as input. The whole arrangement comprises of
highlight arrangement layers, four convolutional and
pooling layers, and completely associated layers. Test
comes about approved with commercial bank online
exchange information appear that the demonstrate gives
great extortion location execution without subordinate
capacities. In expansion, the exactness can be stabilized at
approximately 91% and the review rate at almost 94%, an
increment of 26% and 2%, separately, compared to the
existing negative location CNN [14].
Online trading is a common user behavior with
incredible growth rates today. Completely hassle-free and
becomes a habit of the user. The most common example
can be seen on various e-commerce platforms where more
and more users increase the number of transactions every
day. With such a large online trading network, it is
expected that there will be certain loopholes that make
such platforms fraudulent and conflicting. Therefore, in
this project, we will be discussing a fraud detection and
verification system designed primarily for online
transactions for e-commerce. This document prescribes
certain methods that can be used to authenticate online
transactions and follows verification mechanisms that, to
some extent, help verify the authenticity of transactions if
they turn out to be fraudulent. The aspects described in this
white paper can be looked at in more detail to build a
robust fraud detection system, but they certainly form the
basis for building a viable fraud detection system [15].
Online audit extortion has advanced in
advancement by propelling shrewd campaigns where a
bunch of facilitated members work together to provide
misleading surveys for the assigned targets. Such collusive
extortion is considered much harder to guard against as
these campaign members are able of sidestepping
discovery by forming their behaviors collectively so as not
to seem suspicious. The display work complements
existing things about by investigating more inconspicuous
behavioral trails associated with collusive audit extortion.
A novel factual demonstration is proposed to assist
characterize, recognize, and estimate collusive extortion in
online audits. The proposed demonstration is totally
unsupervised, which bypasses the trouble of manual
comment required for supervised modeling. It is
additionally exceptionally adaptable to incorporate
collusive properties that can be utilized for superior
modeling and forecasting. Tests with two genuine
information sets appear the adequacy of the proposed
strategy and the change of learning and prescient abilities
[15] Fraud in current online advertising activities
increases the risk to online marketing; Advertising industry
and e-business. Click fraud is considered one of the most
important problems on the internet. Advertising. Despite
ongoing efforts by online advertisers to improve traffic
filtering, we are still looking for the best protection to
detect fraudulent clicks. Therefore, effective Online
advertising requires fraud detection algorithms. The
purpose of our article is Determine the accuracy of one of
the latest machine learning algorithms for detecting click
fraud. Online environment. In this study, we studied the
click patterns of a data set of 200 million processing. Click
in 4 days. The main goal is to evaluate the user's click path
on the portfolio and Flags IP addresses that generate a lot
of clicks but don't lead to app installs. As a methodology
we LightGBM - using experimental tests for the Gradient
Boosted Decision Tree Method. this algorithm Provides
98% accuracy. In our study, the literature review was the
central source for testing us result [16]
International Journal of Engineering and Management Research Peer Reviewed & Refereed Journal
e-ISSN: 2250-0758 | p-ISSN: 2394-6962 Volume-13, Issue-4 (August 2023)
https://ijemr.vandanapublications.com https://doi.org/10.31033/ijemr.13.4.6
49
This work is licensed under Creative Commons Attribution 4.0 International License.
While the number of online auctions continues to
grow, the number of online auction scams is also
increasing. To avoid detection, scammers often disguise
normal trading behavior by disguising themselves as
honest participants. So staying vigilant isn't enough to
prevent scams. Online auction participants need a more
proactive approach to protecting their interests, such as an
early fraud detection system. In practice, both accuracy
and timeliness are equally important in developing an
effective detection system. Immediate but misleading
messages to the user are not acceptable. However, a
lengthy discovery process does not help traders place
orders on time. Detection results would be more useful if
potential scammers could be reported as soon as possible.
This study proposes a new fraud early detection method
that considers accuracy and timeliness at the same time. A
modified wrapper procedure was developed to select a
subset of properties from a large pool of candidate
properties to determine the most appropriate properties to
discriminate between regular traders and cheaters. We then
use these properties to propose an additional step-by-step
modeling procedure to extract features from the last part of
a trader's trade history, reducing the time and resources
required for modeling and data collection. You can get
early fraud detection models by building decision trees or
learning from instances. Our experimental results show
that the augmented hybrid model significantly improves
the fraud detection accuracy, while the performance of the
selected attributes is higher than that of the other attribute
sets [17].
Internet banking and e-commerce have
experienced explosive growth over the past few years and
promise tremendous growth in the future. This has made it
easier for scammers to use clever new ways to commit
credit card fraud over the internet. This paper focuses on
real-time fraud detection and presents a new and
innovative approach to understanding spending patterns to
decipher potential fraud cases. Decode, filter, and analyze
customer behavior using self-organizing maps to detect
fraud [18].
This study research attempts to prohibit privacy
and loss of money for individuals and organizations by
creating a reliable model which can detect the fraud
exposure in the online recruitment environments. This
research presents a major contribution represented in a
reliable detection model using ensemble approach based
on Random Forest classifier to detect Online Recruitment
Fraud (ORF). The detection of Online Recruitment Fraud
is characterized by other types of electronic fraud detection
by its modern and the scarcity of studies on this concept.
The researcher proposed a detection model to achieve the
objectives of this study. For feature selection, the support
vector machine method is used and for classification and
detection, ensemble classifier using Random Forest is
employed. This model is applied using a free dataset called
EMSCAD (Employment Scam Aegean Dataset). A
preprocessing step was applied before selection and sorting
were accepted. The result showed an accuracy of 97.41%.
The results also show key features and important factors
for search purposes, including the presence of company
profile features, the presence of company logos and
industry specificities.
E-commerce includes online shopping, banking,
financial institutions, and governments. End Fraud in many
areas of business and everyday life. These activities
Telecom, most common in credit card fraud detection,
network intrusion, finance and insurance and Scientific
applications lose billions of dollars everyday increase in
credit card transactions Use both online and offline at the
same time. to be strong and develop an effective fraud
detection algorithm is the key to reduction. trade loss.
There have been numerous approaches Implemented for
fraud detection. this article shows an approach used for
fraud detection in e-commerce.
With the enhancement in technology e-banking
like credit Card, Debit Card, Mobile Banking, and Internet
Banking is the popular medium to transfer the money from
one account to another. E-Banking is gaining popularity
day by day, which increases the online transaction with the
increase in online shopping, online bill payment like
electricity, Insurance Premium and other charges, online
recharges and online reservation of railways, bus etc., so
the fraud cases related to this are also increasing and it puts
a great burden on the economy, affecting both customers
and financial bodies. It not only costs money, but also a
great amount of time to restore the harm done. The
purpose is to prevent the customer from online transaction
by using specific techniques i.e. based on Data Mining and
Artificial Intelligence technique. The risk score is
calculated by Bayesian Learning Approach to analyze
whether the transaction is genuine or fraudulent based on
the two parameters: Customer Spending Behavior and
Geographical Locations. customers over spending
Behavior you can see using KMEAN clustering Algorithm
and Geolocation Current Your geographic location is
compared to your previous location. A trade is considered
if the risk indicator is greater than 0.5. Fraudulent
Transactions and Security Mechanisms Enter a 4-digit
random number to authenticate the user. appear on the
screen and input by the real user Your code is in the correct
order.
Ⅲ. METHODOLOGY
The methodologies include the algorithm used,
dataset used and flowchart of the data used and
implemented. Below is the provided step by step
explanation of the algorithm used.
International Journal of Engineering and Management Research Peer Reviewed & Refereed Journal
e-ISSN: 2250-0758 | p-ISSN: 2394-6962 Volume-13, Issue-4 (August 2023)
https://ijemr.vandanapublications.com https://doi.org/10.31033/ijemr.13.4.6
50
This work is licensed under Creative Commons Attribution 4.0 International License.
Algorithm Used: The decision tree algorithm is a widely
used supervised learning technique employed for both
classification and regression tasks. It constructs a
structured model resembling a flowchart, driven by input
features.
Tree Construction: The algorithm commences by
considering the entire dataset as the root node, and selects
the optimal feature for partitioning the data.
Feature Split: The chosen feature is utilized to divide the
data into subsets, thereby creating branches or paths within
the decision tree.
Recursive Splitting: The process of feature splitting is
iteratively applied to each subset until a predefined
stopping criterion is satisfied.
Leaf Node Assignment: Leaf nodes are assigned class
labels or regression values based on the majority class or
mean value of the target variable within each respective
subset.
Prediction: To make predictions, the algorithm traverses
the decision tree by evaluating feature values and
ultimately reaching a leaf node to obtain the final
prediction.
Easy to comprehend and interpret
accommodates numerical and categorical data handles
missing values gracefully captures non-
linear relationships effectively.
Prone to over fitting, necessitating proper
regularization techniques - Can be sensitive to changes in
the dataset, leading to instability. Exhibits bias towards
features with high cardinality or many levels In
conclusion, decision trees offer versatility and
transparency in model interpretation. However, caution
must be exercised to address overfitting issues and
effectively manage the algorithm's limitations.
Import necessary libraries and module: from sklearn.
tree import Decision tree classifier, import pandas as pd
and few other libraries are imported.
Read the dataset: use necessary commands to read the
dataset provided and collect the information from it.
Preprocess the data: here the data collected is pre-
processed so that feature extraction can take place easily.
Load the data: The data pre-processed now will be loaded
into the training model to train the model properly.
Splitting the data: The data is splitted into training and
testing data in proportion of 80:20 out of 100.
It checks for the target: if present then goes to next step
or else declares it as unsupervised model as it does not
have specified output.
Checks for target data: we have used the decision tree
classifier therefore uses discrete data.
Training model: after all the necessary adjustments in the
code we train the model using decision tree algorithm.
Testing data: after training the model is tested and
accuracy is obtained with necessary graphs. A small
comparison plot is also included in algorithm.
Results: as last step the results are obtained, and the model
execution ends. Few additional steps of regularization
imputation and many other commands are included in the
code to improve the accuracy and prevent the model from
overfitting and underfitting. This also increases the
generalization of the model and allows it to predict the
new data with more accuracy.
Decision Trees are supervised learning methods
that can be used to solve classification and regression
issues, however they are typically chosen for Classification
issues. It is a tree-structured classifier, with internal nodes
denoting dataset features, branches denoting decision
rules, and each leaf node denoting the classification result.
There are two nodes—the Decision Node and the Leaf
Node—in a decision tree. Decision nodes are used to make
decisions and have many branches, whereas Leaf nodes
represent the results of those decisions and do not have any
additional branches. Based on the characteristics of the
available dataset, judgments or tests are run.
Here is a brief explanation of the flowchart of
model training in figure 3.
1. Start: The flowchart begins with the start symbol,
indicating the beginning of the decision tree algorithm.
2. Load Dataset: The algorithm loads the dataset, which
contains the input features and target variable.
3. Define Features and Target: The feature columns and
target column are defined, specifying the variables to be
used for training the decision tree.
4. Split Data: The dataset is split into training and testing
sets using the train_test_split function, allocating a portion
of the data for model evaluation.
5. Data Imputation: The SimpleImputer object is used to
handle missing values in the dataset, replacing them with
the mean value of the respective feature.
6. Build Decision Tree: The DecisionTreeClassifier object
is created, representing the decision tree model. It is
trained on the training data using the fit function.
7. Predictions: The trained decision tree is utilized to
make predictions on the test set, using the predict function.
International Journal of Engineering and Management Research Peer Reviewed & Refereed Journal
e-ISSN: 2250-0758 | p-ISSN: 2394-6962 Volume-13, Issue-4 (August 2023)
https://ijemr.vandanapublications.com https://doi.org/10.31033/ijemr.13.4.6
51
This work is licensed under Creative Commons Attribution 4.0 International License.
IV. FLOWCHART
Figure 3: Flowchart of model training
8. Evaluate Accuracy: The accuracy of the model is
calculated by comparing the predicted values with the
actual target values using the accuracy score function.
9. Display Results: The accuracy score is printed to the
console, providing an assessment of the model's
performance.
10. End: The flowchart concludes with the end symbol,
indicating the completion of the decision tree algorithm.
The flowchart shown in figure 3 provides a visual
representation of the steps involved in training and
evaluating the decision tree model, aiding in understanding
the overall process and facilitating communication
between different stakeholders.
Datasets: The dataset currently used in the model training
is taken from the Kaggle website. Kaggle is a popular
platform for data scientists and machine learning
practitioners to discover and share datasets, as well as
participate in data science competitions. It hosts a vast
collection of datasets from various domains, allowing
users to access and analyse real-world data. Kaggle
datasets are typically provided in structured formats such
as CSV, Excel, or SQL, and may contain a wide range of
variables or features. These datasets cover diverse topics
including healthcare, finance, social sciences, computer
vision, natural language processing, and more. Users can
explore and download datasets from Kaggle for their own
analysis, model training, or research purposes. They can
also contribute by uploading and sharing their own
datasets with the Kaggle community. Kaggle datasets are
accompanied by detailed descriptions, documentation, and
often include pre-split train/test datasets to facilitate model
development and evaluation. Additionally, many datasets
come with sample code notebooks, known as kernels,
which provide examples and insights on how to work with
the data. By leveraging the vast collection of Kaggle
datasets, data scientists can gain access to high-quality,
real-world data to explore, analyse, and build predictive
models or develop insights for a wide range of
applications.
Ⅴ. RESULTS AND DISCUSSIONS
The machine learning model utilizes python code
in Jupyter notebook to train and test the model. The model
uses decision tree algorithm. The code starts from training
the feature and target columns from dataset. Numpy is a
Python library for efficient numerical computations,
offering multidimensional arrays and mathematical
functions. It is widely used in scientific computing and
data analysis. Pandas is a powerful data manipulation
library built on top of Numpy. It provides high-level data
structures like Data Frames and Series, making data
manipulation and analysis easier. Data Frames are two-
dimensional tables with labelled rows and columns, while
Series are one-dimensional labelled arrays. Pandas offers
tools for data cleaning, pre-processing, merging,
reshaping, and analysing structured data. It supports
flexible indexing, filtering, and grouping operations,
allowing easy extraction and manipulation of specific
subsets of data. Numpy and Pandas are widely used in
scientific computing, data analysis, and machine learning
applications. Numpy provides efficient storage and
manipulation of large arrays, while Pandas provides
intuitive data manipulation capabilities. Both libraries
integrate well with other Python data ecosystem tools like
Matplotlib and Scikit-learn. Numpy and Pandas are
essential for data manipulation, analysis, and
preprocessing in Python. Together, they provide efficient
and convenient tools for working with arrays and
structured data. The first code snippet used in code in
summary, the code performs data preprocessing, splits the
data into training and test sets, trains two Decision Tree
Classifiers with different parameters, and evaluates their
accuracy in predicting the target variable. Here a simple
imputer is included. The Simple Imputer is a class from the
scikit-learn library that provides a simple strategy for
handling missing values in a dataset. It is used to replace
International Journal of Engineering and Management Research Peer Reviewed & Refereed Journal
e-ISSN: 2250-0758 | p-ISSN: 2394-6962 Volume-13, Issue-4 (August 2023)
https://ijemr.vandanapublications.com https://doi.org/10.31033/ijemr.13.4.6
52
This work is licensed under Creative Commons Attribution 4.0 International License.
missing values with a chosen strategy, such as the mean,
median, or most frequent value of the respective feature.
In summary, next code snippet utilizes a trained Decision
Tree Classifier to generate a visual representation of the
decision boundaries in the feature space. The decision
boundaries are plotted as filled contour regions, and the
training points from the dataset are displayed with colours
representing different features. This visualization helps in
understanding how the classifier separates and categorizes
the data points based on their feature values. Decision
boundaries are lines or surfaces that separate different
classes in a classification problem. They define the regions
where a classifier assigns different labels. They are
determined by learning from the training data and finding
the optimal separation between classes. Decision
boundaries are influenced by the features used for
classification and can vary in complexity. Visualizing
decision boundaries helps understand how a classifier
distinguishes classes and identifies areas of uncertainty.
The complexity of decision boundaries reflects the
relationships between features. They are crucial for
evaluating classifiers and assessing their generalization
ability. Decision boundaries aid in model interpretation,
evaluation, and decision-making, providing a visual
representation of class separation in the feature space it is
shown in figure 4.
Figure 4: Decision boundaries in feature space
[time=black colour, id=red colour]
The distribution of transaction types in the dataset
is depicted using a pie chart in the given code. The
generated pie chart gives you a general idea of how
different transaction types are distributed throughout the
dataset and enables you to see how many of each type
there are. The size of each pie slice shows the relative
frequency or occurrence of a certain transaction type in the
dataset, and each slice represents a particular transaction
type. The use of logarithmic trends in data analysis and
visualization is to better understand and represent
exponential or multiplicative relationships between
variables. By plotting data on a logarithmic scale, it
compresses large ranges of values and magnifies smaller
changes, making it easier to observe patterns and trends. In
the provided code, the logarithmic trends of different
features in the dataset are visualized using line plots,
allowing for a clearer understanding of the relationships
and patterns present in Figure 5.
Figure 5: Pie Chart
The code employments the seaborn library
(imported as sns) and matplotlib.pyplot (imported as plt) to
make and show a histogram. The plt.figure (fig size=(10,
6)) line sets the measure of the figure, guaranteeing the
coming about plot has appropriate
dimensions.sns.histplot() is utilized to form the histogram.
The information parameter is set to the Data Frame df,
demonstrating that the 'amount' column will be utilized for
the histogram. The x parameter indicates the column to be
plotted. The kde=True contention includes a part thickness
estimation plot to the histogram.Additional customizations
are made utilizing matplotlib. plt.title () sets the title of the
plot as 'Histogram of Exchange Amount'. plt.xlabel () and
plt.ylabel () are utilized to name the x-axis and y-axis,
respectively.Finally, plt.show() is called to show the
histogram. The coming about histogram provides a visual
representation as in figure 6, of the dissemination of
exchange sums within the dataset. The x-axis speaks to the
exchange sum, and the y-axis speaks to the recurrence of
exchanges falling inside each sum run. The histogram
makes a difference to recognize designs, exceptions, and
the in general shape of the dispersion, giving bits of
knowledge into the exchange sums show within the
dataset. Figure 6 represents the expansion of the bit
thickness estimation plot (kde=True) gives a smoothed
representation of the dissemination, permitting for way
better visualization of the basic thickness as.
International Journal of Engineering and Management Research Peer Reviewed & Refereed Journal
e-ISSN: 2250-0758 | p-ISSN: 2394-6962 Volume-13, Issue-4 (August 2023)
https://ijemr.vandanapublications.com https://doi.org/10.31033/ijemr.13.4.6
53
This work is licensed under Creative Commons Attribution 4.0 International License.
Figure 6: Histogram
The code to begin with calculates the relationship
framework by calling df.corr() on the DataFrame df. The
relationship lattice speaks to the pairwise relationships
between the numeric columns within the dataset.
Next, the code makes a heatmap utilizing
seaborn's heatmap() work. The annot=True parameter
permits showing the relationship values on the heatmap,
and the cmap='coolwarm' parameter sets the color plot for
the heatmap.
We set the figure measure utilizing
plt.figure(figsize=(10, 8)) to guarantee the heatmap has
fitting dimensions.
Additional customizations are made utilizing
matplotlib (imported as plt). We set the plot title utilizing
plt.title().
Finally, plt.show() is called to show the heatmap.
The coming about heatmap visualizes the
correlations between factors within the dataset. Positive
relationships are demonstrated by hotter colors (e.g.,
ruddy), negative relationships by cooler colors (e.g., blue),
and no relationship by a neutral color (e.g., white).
By analyzing the heatmap, you'll recognize
connections between factors, such as solid positive or
negative relationships, which can give experiences into the
dataset's designs and conditions. As shown in figure 7.
Figure 7: Correlation Heatmap
Begin at the best hub (root hub) of the tree. Each
hub speaks to a choice based on a highlight and a edge
value. Based on the highlight and limit esteem at each
hub, take after the fitting department. On the off chance
that the condition is fulfilled, move to the cleared out child
hub; something else, move to the correct child node.
Continue navigating the tree until you reach a leaf
node. Leaf hubs speak to the ultimate choice or prediction.
The color of the hubs within the chart demonstrates the
lesson dispersion. Filled hubs have a larger part of tests
having a place to a particular class.
The number in each hub speaks to the lesson
dispersion and the check of tests for each class.
By analysing the choice tree chart, you'll get it the
choice rules utilized by the calculation and pick up bits of
knowledge into how distinctive highlights contribute to the
classification process.
Make beyond any doubt to alter the variable
names (demonstrate, xtrain, ytrain, x.columns) agreeing to
your particular code and information.
In order to understand the behaviour of the
features with respect to each other scatter plots are plotted.
The code uses the Pandas and Plotly libraries to create a
subplot with 6-line plots representing different features of
International Journal of Engineering and Management Research Peer Reviewed & Refereed Journal
e-ISSN: 2250-0758 | p-ISSN: 2394-6962 Volume-13, Issue-4 (August 2023)
https://ijemr.vandanapublications.com https://doi.org/10.31033/ijemr.13.4.6
54
This work is licensed under Creative Commons Attribution 4.0 International License.
a dataset. The dataset is loaded using Pandas, and a subplot
with 6 rows and 1 column is created using Plotly's `make
subplots () `.
The output of the code is a countplot that displays
the distribution of the 'target' column. Each bar represents
the count of occurrences for each unique value in the
'target' column. The countplot provides insights into the
class distribution and can be used to analyze the balance or
imbalance of classes in the dataset.
The decision tree is shown visually as the code's
output. The arrows show the flow from one node to
another based on the decisions, and each node in the tree
reflects a decision based on a feature. The majority class
for each node is represented by the node's color. The
decision-making process of the decision tree classifier
based on the characteristics and their significance in
forecasting is depicted visually in the plot.
In summary, next code loads a dataset using
Pandas, separates the features and the target variable, and
creates an instance of the DecisionTreeClassifier from
scikit-learn. It then fits the classifier on the dataset and
plots the decision tree using the tree.plot_tree() function.
The resulting plot visualizes the decision tree structure,
with each node representing a decision based on a specific
feature, and the leaf nodes indicating the predicted classes.
The plot is displayed using matplotlib. pyplot. show (). A
decision tree plot is a graphical representation that
illustrates the decision-making process of a decision tree
classifier or regressor. It visually depicts the hierarchical
structure of the decision tree, where nodes represent
decision points and branches represent possible outcomes.
The plot displays the feature names and their
corresponding thresholds at each decision node. Terminal
nodes, or leaves, signify the final decisions or predicted
outcomes. The size and colour of the nodes can be utilized
to convey the sample count or class distribution at each
node. The plot offers a clear and intuitive representation of
the decision tree's logic and decision boundaries. It aids in
comprehending how the decision tree makes predictions
based on various features and thresholds. Figure 8, The
plot can be customized to include class labels, feature
importance, and other pertinent information. It is valuable
for interpreting and explaining the decision tree model to
stakeholders or non-technical audiences. The decision tree
plot allows for easy identification of significant features
and their influence on the decision-making process. It
assists in identifying regions of high predictive accuracy as
well as areas where the model may struggle to make
accurate predictions. The plot can be generated using
various Python libraries, such as scikit-learn, matplotlib, or
plotly. It helps identify potential overfitting or underfitting
issues by visualizing the complexity of the decision tree.
The plot serves as a valuable tool for model validation,
comparison, and explanation. Figure 9. It facilitates
effective communication between data scientists, analysts,
and stakeholders by providing a visual representation of
the decision tree's logic. The decision tree plot is a
powerful visualization tool that enhances the
interpretability and transparency of decision tree models.
Plot has a max depth of 2 as shown in Max depth is
command used to reduce the complexity of model and
improve model accuracy. Figure 10.
Figure 8: Transaction Distribution Graph
International Journal of Engineering and Management Research Peer Reviewed & Refereed Journal
e-ISSN: 2250-0758 | p-ISSN: 2394-6962 Volume-13, Issue-4 (August 2023)
https://ijemr.vandanapublications.com https://doi.org/10.31033/ijemr.13.4.6
55
This work is licensed under Creative Commons Attribution 4.0 International License.
Figure 9: Logarithmic trend, oldbalanceOrg
Figure.10: Logarithmic trend, newbalanceOrg
International Journal of Engineering and Management Research Peer Reviewed & Refereed Journal
e-ISSN: 2250-0758 | p-ISSN: 2394-6962 Volume-13, Issue-4 (August 2023)
https://ijemr.vandanapublications.com https://doi.org/10.31033/ijemr.13.4.6
56
This work is licensed under Creative Commons Attribution 4.0 International License.
By looking at the outputs we can say that for this
application regression tree is the best fit. The code analyses
the "archive.csv" dataset by performing various data
analysis and visualization tasks. Firstly, the dataset is
loaded using pandas. Then, violin plots are created to
visually explore the distribution of features in relation to
the target variable. A correlation heatmap is generated to
investigate the relationships between features and the
target variable. The data is prepared for training by
splitting it into training and testing sets. Missing values are
handled through data imputation using a simple imputer. A
decision tree classifier is trained on the training data.
Predictions are made on the test set, and the accuracy score
is calculated to evaluate the classifier's performance. A
confusion matrix is created to visualize the classification
results. Finally, a table visualization is generated to
provide an overview of the dataset. Overall, this code
provides valuable insights into feature distributions,
correlations, and the effectiveness of the decision tree
classifier for predicting the target variable. Figure 11.
A classification report summarizes the performance of a
classification model on a dataset, providing an evaluation
of its predictive accuracy for each class. It includes metrics
such as precision, recall, F1-score, and support for each
class. Precision measures the accuracy of positive
predictions, while recall evaluates the model's ability to
find positive instances. The F1-score combines precision
and recall into a balanced measure of performance.
Support indicates the number of instances for each class in
the dataset. The classification report helps assess a model's
strengths and weaknesses in classifying different classes,
making it useful for comparing algorithms or different
configurations.
Table 1: Model performance
Figure 11: Correlation Matrix
Ⅵ. CONCLUSION AND FUTURE
SCOPE
An online fraud detection system implemented
using a decision tree algorithm shows promising results in
detecting fraudulent transactions. A satisfactory accuracy
score was obtained by analysing various characteristics
such as transaction type, amount, old balance, and new
balance. Decision tree algorithms are efficient at capturing
complex patterns and making decisions based on feature
values. Subdivide the data set based on the most
informative features to create a tree structure for easier
decision making. This algorithm has the advantage of
being interpretable because the decision rules are easy to
understand and visualize. Future Scope: Feature
Engineering: Exploring additional features or transforming
existing features may improve the model's performance.
Consider incorporating time-based features, behavioural
patterns, or derived features from domain knowledge.
Ensemble Methods: Implementing ensemble methods such
as Random Forest or Gradient Boosting can potentially
enhance the model's accuracy and robustness. Ensemble
methods combine multiple decision trees to make
collective predictions, reducing overfitting and increasing
overall performance. Anomaly Detection: Incorporating
anomaly detection techniques alongside the Decision Tree
algorithm can enhance fraud detection capabilities.
Unsupervised learning methods like clustering or outlier
detection can help identify suspicious patterns or outliers
in transaction data. Real-time Monitoring: Integrating the
model into a real-time monitoring system can enable
continuous fraud detection and prevention. By
implementing a streaming data pipeline and leveraging
technologies like Apache Kafka or Apache Flink, the
system can analyse incoming transactions in real-time.
Improved Data Collection: Collecting more diverse and
representative datasets can help enhance the model's
generalization and effectiveness. Obtaining labelled data
International Journal of Engineering and Management Research Peer Reviewed & Refereed Journal
e-ISSN: 2250-0758 | p-ISSN: 2394-6962 Volume-13, Issue-4 (August 2023)
https://ijemr.vandanapublications.com https://doi.org/10.31033/ijemr.13.4.6
57
This work is licensed under Creative Commons Attribution 4.0 International License.
for both fraudulent and non-fraudulent transactions is
crucial to improve the model's ability to accurately detect
fraud. model evaluation. Thorough evaluation and
validation of model performance on a variety of data sets
and the use of appropriate evaluation metrics can provide a
comprehensive assessment of performance. Consider
testing the model's performance in different scenarios
using methods such as cross-validation or A/B testing.
Continuously updating and improving fraud detection
systems, incorporating advanced algorithms, and
continually updating new fraud models and methods are
critical to ensuring effective fraud prevention in online
transactions.
ABBREVIATIONS
Online Fraud Detection (OFD)
Decision tree (DT)
False positive (FP)
False negative (LN)
True Positive (TP)
True Voice (TN)
Accuracy (ACC)
Accuracy (PR)
Recall (REC)
F1 score (F1)
REFERENCES
[1] Abdallah, Aisha, Mohd Aizaini Maarof &
Anazida Zainal. (2016). Fraud detection system:
A survey. Journal of Network and Computer
Applications, 68, 90-113.
[2] Hu, Nan, Ling Liu & Vallabh Sambamurthy.
(2011). Fraud detection in online consumer
reviews. Decision Support Systems, 50(3), 614-
626.
[3] Minastireanu, Elena-Adriana & Gabriela Mesnita.
(2019). An analysis of the most used machine
learning algorithms for online fraud
detection. Informatica Economica, 23(1).
[4] Zhang, Zhaohui, et al. (2018). A model based on
convolutional neural network for online
transaction fraud detection. Security and
Communication Networks.
[5] Akoglu, Leman, Rishi Chandy & Christos
Faloutsos. (2013). Opinion fraud detection in
online reviews by network effects. Proceedings of
the International AAAI Conference on Web and
Social Media, 7(1).
[6] Chauhan, Nidhika & Prikshit Tekta. (2020). Fraud
detection and verification system for online
transactions: a brief overview. International
Journal of Electronic Banking, 2(4), 267-274.
[7] Xu, Chang & Jie Zhang. (2015). Towards
collusive fraud detection in online reviews. IEEE
International Conference on Data Mining.
[8] Minastireanu, Elena-Adriana & Gabriela Mesnita.
(2019). Light gbm machine learning algorithm to
online click fraud detection. J. Inform. Assur.
Cybersecur, 263928.
[9] Chang, Wen-Hsi & Jau-Shien Chang. (2012). An
effective early fraud detection method for online
auctions. Electronic Commerce Research and
Applications, 11(4), 346-360.
[10] Kewei, Xiong, et al. (2021). A hybrid deep
learning model for online fraud detection. IEEE
International Conference on Consumer
Electronics and Computer Engineering.
[11] Zhang, Ruinan, Fanglan Zheng & Wei Min.
(2018). Sequential behavioral data processing
using deep learning and the Markov transition
field in online fraud detection. arXiv preprint
arXiv:1808.05329.
[12] Chang, Wen-Hsi & Jau-Shien Chang. (2012). An
effective early fraud detection method for online
auctions. Electronic Commerce Research and
Applications, 11(4), 346-360.
[13] Cao, Shaosheng, et al. (2019). Titant: Online real-
time transaction fraud detection in ant
financial. arXiv:1906.07407.
[14] AC, Ramachandra & Venkata Siva Reddy. (2022).
Bidirectional DC‑DC converter circuits and
smart control algorithms: A review.
[15] Kumari, Ashwini, et al. (2018). Multilevel home
security system using arduino & gsm. Journal for
Research 4.
[16] Viswanatha, V., et al. (2020). Intelligent line
follower robot using MSP430G2ET for industrial
applications. Helix-The Scientific Explorer,
10(02), 232-237.
[17] Viswanatha, V. & R. Reddy. (2020).
Characterization of analog and digital control
loops for bidirectional buck–boost converter
using PID/PIDN algorithms. Journal of Electrical
Systems and Information Technology, 7(1), 1-25.
[18] Viswanatha, V., R. K. Chandana & A. C.
Ramachandra. (2022). IoT based smart mirror
using raspberry pi 4 and yolo algorithm: A novel
framework for interactive display.
