Unfolding the Credit Card Fraud Detection Technique by Implementing SVM Algorithm

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 01 | Jan 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 942
Unfolding the Credit Card Fraud Detection Technique by
Implementing SVM Algorithm
Bhavesh Gujar1, Ankush Ginjari2, Sushant Phase3, Ashutosh Singh4, Prof. Chitralekha Dwivedi5
1Student, Dept. of Information Technology, Dr. D.Y. Patil Institute of Technology, Pune, Maharashtra, India
5Assistant Professor, Dept. of Information Technology, Dr. D.Y. Patil Institute of Technology, Pune, Maharashtra,
India
------------------------------------------------------------------------***-------------------------------------------------------------------------
Abstract - Credit card fraud is the most prevalent problem
in today's society, and it must be addressed immediately.
"The act of washing filthy money such thatthesourceofcash
can no longer be identified is known as credit card fraud."
Credit card fraud is difficult to detect since massive financial
transactions involving large sums of money occur on a
regular basis in the global market. The (Anti-credit card
fraud Suite) was introduced to detect suspicious activity,
although it only applies to individual transactions, not bank
account transactions, as previously claimed. We describe a
machine learning method based on 'Structural Similarity,'
which finds common qualities and behaviours among bank
account transactions to address these issues. Due to the
difficulties of detecting credit card fraud transactions from
large datasets, we provide case reduction algorithms that
reduce the input dataset beforelocatingpairsoftransactions
with other bank accounts that have comparable attributes
and behaviours.
Key Words: SVM, Machine Learning, pre-processing,
Classification, deep learning.
1. INTRODUCTION
Credit is used to obtain a loan and then repay it over a set
period of time. Credit ratings are used to determine how
likely a person is to repay their debts, and they must be
utilised in order to give money to someone who may be a
debtor. [1]. When credit scoringisemployed,itiscritical that
the information regarding debtors be appropriately
classified. You must perform a math computationinorder to
extract relevant information from the data. Data mining is a
discipline of science that aims to extract useful data and
knowledge from massive datasets. [2]. It is one of the data
mining approaches for categorising items into groups.
Things are classified in a variety of ways, including Decision
Trees and Support Vector Machines. This research will
employ the C4.5 algorithm. Quinlan (1996) proposed the
C4.5 algorithm as an improved version of the ID3 algorithm.
Only categorical (nominal or ordinal) type features can be
used to create the decision tree in ID3. However, numerical
types can be utilised to construct the tree (intervalsorratios
can not be used). This is an example of how the
modifications to ID3 in C4.5 enable it to handle numeric
types.
Credit receipts are created by categorising debtors intotwo
groups: those who have good credit and those who have
negative credit. [4]. In thisinvestigation,creditdatasetsfrom
Germany were used (GDC). A dataset is a collection of items
and their associated properties. The featuresofanobjectare
the characteristics that distinguish it from others. [5] The
German Credit Dataset is available in the UCI Machine
Learning Repository. This data is being worked on by Dr.
Hans Hofmann, a professor at the University of Hamburg.
There are 20 different features, 1000 different occurrences,
and two different credit categories in the German Credit
Datasets.
The purpose of this work is to increase the C4.5 algorithm's
accuracy. The primary purpose of this research is to reduce
GCD features. This paper proposed feature distribution and
feature splits in datasets. In the preprocessing stage, split
feature reduction is used. It's employed since GCD includes
20.000 data points, each with 20 features and 1000
inferences. To extract the best features and improve
accuracy, feature reduction is done [6]. Bagging Ensemble
[7] is also used to select a suitable ensemble for the C4.5
algorithm. Muslim et al. discoveredthatbyusingtheBagging
Ensemble, the C4.5 approach can be improved. People will
compare the C4.5 algorithm to how it would work onitsown
to evaluate how this inquiry compares.
2. PROBLEM STATEMENT
The purpose of this project is to identify illegal credit card
and bank transaction activity. In the early stages of
development, it is still in its youth. To decrease the level of
criminal activities. Transmitting funds for purposes other
than what they were designed appears to be a common
practise in the Credit Card industry.
3. LITERATURE SURVEY
Hongwei Chen1 , He Ai , Zhihui Yang1 , Weiwei Yang1 ,
Zhiwei Ye1 , Dawei Dong,” An Improved XGBoost Model
Based on Spark for Credit Card Fraud Prediction.’’[1] Many

financial organisations suffer significant losses due to credit
card theft. Credit card fraud data is so dense that an updated
XGBoost model built on Spark is neededtodeal withthedata
imbalance. Balanced training sets were created using the
Smote algorithm. The fraud detection system made use of
the Spark-based XGBoost classifier. After that, the test sets
were sorted out one by one. As part of a model comparison
experiment, we compared this project's XGBoost algorithm
to other algorithms such as logistic regression and decision
trees as well as random forests. There is a 9.1 percent
difference between the model provided in this project and
the model rated second in terms of Recall, F1-ScoreandAUC
in the trial data. The speedup on the datasets of 70,000,
140,000, and 280,000 samples is 2.06, 3.28, and 3.75,
respectively, in the experiment with increased throughput.
Experiment findings show that the proposed model
accurately and efficiently predicts credit card theft andhasa
practical effect.
Bing Zhu, Wenchuan Yang, Huaxuan Wang, Yuan Yuan,” A
Hybrid Deep Learning Model for Consumer Credit
Scoring”[2] Consumer credit scoring is an essential part of
data mining for risk management of consumer credit, and a
variety of data mining methodologies have been created or
used to it. Image recognition, computer vision, and other
fields have experienced a rise in the use of deep learning
techniques in recent years. In this essay, we are using deep
learning to improve credit ratings for consumers. The
procedure of deciding on features Using a convolutional
neural network and relief, we've created a new hybrid
model. In studies done on a real-world dataset from a local
consumer loan organisation, theproposedmodel beatsother
benchmark models like logistic regression and random
forest.
Much Aziz Muslim, Aldi Nurzahputra, Budi Prasetiyo“
Improving Accuracy of C4.5 Algorithm Using Split Feature
Reduction Model and Bagging Ensemble forCreditCardRisk
Prediction”[3] Whether or not a prospective consumer is
granted credit is determined by the existence of credit
scoring. To accurately classifya debtor,creditscoresmustbe
accurate enough. It is possible to classify data in various
ways, including the decision tree. Decision tree algorithms
such as the C4.5 algorithm can be used. This study aims to
improve the C4.5 algorithm's ability to anticipate credit
receipts. Accuracy is improved with the Split Feature
Reduction Model and the Bagging Ensemble. It is used in the
pre-processing step to divide datasetsinthenumberofparts
n. Four categories of data were used to create this article.
There are 16 features in Split 1; 12 features in Split 2; 8
features in Split 3; and 4 features in Split 4. The C4.5
algorithm is then applied to each split. The C4.5 algorithm
and the split feature reduction model yielded an accuracy
rate of 73.1% in Split 3. 75.1 percent in Split 3 is the best
accuracy achieved by employing the split feature reduction
model and bagging ensemble with the C4.5 algorithm. The
combined use of a split feature reduction model and a
bagging ensemble improved accuracy by 4.6% over the C4.5
algorithm alone.
MILLER ARIZA1,2, JAVIER ARROYO1,3, ANTONIO
CAPARRINI4, and MARÍA-JESÚS SEGOVIA.” Explainabilityof
a Machine Learning Granting Scoring Model in Peer-to-Peer
Lending”[4] Peer-to-peer lending necessitates credit risk
models that are both efficient and transparent. Traditional
machine learning algorithms perform well in terms of
prediction but fall short in terms of explanatory capabilityin
the vast majority of cases. However, recent explainability
methods, such as the SHAP values, can be employed to
circumvent this problem. In this work, we test the well-
known logistic regressionmodel againsta varietyofmachine
learning algorithms for assigning score in peer-to-peer
lending. According to the comparison, the machine learning
option outperforms in terms ofclassificationperformance as
well as explainability. SHAP values, in particular,
demonstrate how machine learning can account for
dispersion, nonlinearity, and structural fractures in feature-
target variable interactions. Our findings indicate that
machine learning can be given credit scoringmodelsbe both
accurate and transparent.
Mary Frances Zeager, Aksheetha Sridhar, Nathan Fogal,
Stephen Adams, Donald E. Brown, and Peter A. Beling,”
Adversarial Learning in Credit Card Fraud Detection”[5]
Credit card theft is a severe problem for many financial
institutions, costing them billions of dollars each year.
Because fraud detection methods do not incorporate
information on the adversary's understanding of the fraud
detection mechanism, many adversaries continue to avoid
detection. The purpose of this research is to develop a
dynamic fraud detection system that incorporates
information about the "fraudster's" objectives and
knowledge base. We employ a game theoretical adversarial
learning technique in this research to model a fraudster's
best strategy and adjust the fraud detection system ahead of
time to better identify future fraud cases. Using a logistic
regression classifier as the fraud detection method, we first
evaluate the enemy's optimal approachbasedontheamount
of undetected fraud cases.
4. PROPOSED SYSTEM
This section goes into great detail about the proposed ML
framework. The framework is constructedinsuchawaythat,
given a collection of bank accounts and transactions, it will
provide a list of probable money laundering account groups.
The framework begins by examining the input and looking
for transactions that match. Matching transactions are
financial transactions that share characteristics such as
deposit and withdrawal amounts. Following that, the
framework constructs a network representation of all
matching transactions. It then usenetwork-basedalgorithms
to eliminate redundant accounts and transactions. The
approach then uses a clustering algorithm to identify
questionable ML communities in the network. The term
"proposed system" refers to all components, including
hardware and software, included in the Respondent's
proposal. The Proposed System should include Document
Management, Workflow, and Customer Relationship

Management framework's efficiency. Then, using structural
similarities, we can identify and group possible credit card
accounts. Our preliminaryexperimentalresultsshowthatwe
can detect ML accounts with a high degree of accuracy.The
prior information on the dataset, such as its attributes,
dimensions, and data types of each feature, etc., is an
essential factor that helps one to perform proper operations.
An offline dataset which is a publically accessible web
platform named “Kaggle” is considered for the
implementation of the program. The dataset is a Credit card
fraud dataset that consists of several transactions. The
dataset contains a combination of cases of fraud and non-
fraud. CSV files are the most commonly used format for
machine learning data. The dataset contains rows and
columns of the following features like Merchant_id,
Transaction amount, Is declined, Total Number of declines
per day, is Foreign Transaction, is HighRisk Country, and is
Fraudulent.
Figure 1. System Architecture
5. ALGORITHM
SVM (Support Vector Machine)
The hyperplane (basically a two-dimensional line) that
involves different the tags is created by a support vector
machine using these data points. This connect as a decision
boundary: anything on one side is categorized as blue, while
everything on the other is classifiedasred. Thesedata points
are used by a support vector machine to build a hyperplane
(basically a two-dimensional line) with various tags. This
link serves as a decision boundary: anything on one side is
labelled blue, while everything on the other side is labelled
red. SVMs are supervisedmachinelearningmethodsthatcan
be used to tackle classification and regression problems. It
alters your data using a technique known as the kernel trick
before determining a suitable border between the available
outputs based on these adjustments. The goal of the SVM
algorithm is to find a hyper - plane in an N-dimensional
space that categorises the input points clearly. The number
of features determines the size of the hyperplane. The
hyperplane is essentially a line if there are only two input
features. The goal of the SVM algorithm is used to find a
hyperplane in an N-dimensional space that categorises the
input points clearly. The number of features determines the
size of the hyperplane. The hyperplane is essentially a line if
there are only two input features. SVM can be employed
when the number of features in the dataset is large in
contrast to the number of data points. By using the
appropriate kernel and configuring the best set of
parameters. SVM is an excellent, but not the best, classifier.
In truth, no one can claim to be the best. SVM works
reasonably effectively when there is a clear margin of
distinction.
6. CONCLUSION
The proposed machine learning system aims to identify
possible Credit card fraud groups within a huge number of
economic transactions. Case reducing methods such as
matching transaction detection and balance score filter are
used to restrict the list of prospective ML accounts in order
to improve theframework'sefficiency.Then,usingstructural
similarities, we can identify and group possible credit card
accounts. Our preliminary experimental results show that
we can detect ML accounts with a high degree of accuracy.
REFERENCES
[1] Rokach, Lior and Maimon, Oded. Data Mining with
Decision Trees: Theory and Applications 2nd Eddition.
Singapore: World Scientific Publishing Co. 2015.
[2] Sugiharti, E. and Muslim, M.A.,. “On-Line Clustering of
Lecturers Performance of Computer Science Department of
SemarangStateUniversityUsingK-Meansalgorithm”. Journal
of Theoretical and Applied Information Technology, 83(1),
p.64. 2016
[3] Prasetyo, Eko. Data mining: Mengolah Data menjadi
Informasi Menggunakan MATLAB. Yogyakarta:AndiOffsett.
2014.
[4] Nurzahputra, A. and Muslim, M.A., “Peningkatan Akurasi
Pada Algoritma C4. 5 Menggunakan Adaboost Untuk
Meminimalkan Resiko Kredit”. Prosiding SNATIF, pp.243-
247. 2017.
[5] Hermawati, F. A. DATA MINING. Yogyakarta: Andi Offset.
2013.
[6] Wijaya, K. P. & Muslim, M. A. “Peningkatan Akurasi Pada
Algoritma Support Vector Machine Dengan Penerapan
Information Gain Untuk Mendiagnosa Chronic Kidney
Disease”. Prosiding 3rd Seminar Nasional Ilmu Komputer.
Semarang: Universitas Negeri Semarang. 2016.
[7] Lessmann, S., Baesens, B., Seow, H.V., and Thomas, L.C.,
2015. Benchmarking state-of-the-art classification

algorithms for credit scoring: An update of research.
European Journal of Operational Research 247, 1, 124-136.
[8] Crook, J.N., Edelman, D.B., and Thomas, L.C.,2007.Recent
developments in consumercreditrisk assessment.European
Journal of Operational Research 183, 3, 1447-1465.
[9] Z. Kazemi and H. Zarrabi, "Using deep networks forfraud
detection in the credit card transactions," 2017 IEEE 4th
International Conference on Knowledge-Based Engineering
and Innovation (KBEI), Tehran, 2017, pp. 0630-0633.
[10] D. Prusti and S. K. Rath, "Web service based credit card
fraud detection by applying machine learning techniques,"
TENCON 2019 - 2019 IEEERegion10Conference(TENCON),
Kochi, India, 2019, pp. 492-497.
[11]Rokach, Lior and Maimon, Oded. Data Mining with
Decision Trees: Theory and Applications 2nd Eddition.
Singapore: World Scientific Publishing Co. 2015.
[12] Sugiharti, E. and Muslim, M.A.,. “On-Line Clustering of
Lecturers Performance of Computer Science Department of
SemarangStateUniversityUsingK-Meansalgorithm”. Journal
of Theoretical and Applied Information Technology, 83(1),
p.64. 2016.
[13] M. Leo, S. Sharma, and K. Maddulety, “Machinelearning
in banking risk management: A literature review,”Risks,vol.
7, no. 1, 2019.
[14] C. Serrano-Cinca and B. Gutierrez-Nieto, “The use of
profifit scoring as an alternative to credit scoring systems in
peer-to-peer (P2P) lending,” Decision Support Systems, vol.
89, pp. 113–122, 2016.

Unfolding the Credit Card Fraud Detection Technique by Implementing SVM Algorithm

Recommended

More Related Content

Similar to Unfolding the Credit Card Fraud Detection Technique by Implementing SVM Algorithm (20)

More from IRJET Journal (20)

Recently uploaded (20)

Unfolding the Credit Card Fraud Detection Technique by Implementing SVM Algorithm