See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/336552027

A Review On Credit Card Fraud Detection Using Machine Learning

Article  in  International Journal of Scientific & Technology Research · October 2019

CITATIONS READS
0 576

4 authors, including:

Chetan J Awati Rashmi More

Shivaji University, Kolhapur Shivaji University, Kolhapur
44 PUBLICATIONS   21 CITATIONS    1 PUBLICATION   0 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Overcome Key Escrowing problem with attribute based access control policy & efficient cloud environment View project

Credit card Fraud detection View project

All content following this page was uploaded by Chetan J Awati on 15 October 2019.

The user has requested enhancement of the downloaded file.

 INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 8, ISSUE 10, OCTOBER 2019 ISSN 2277-8616

A Review On Credit Card Fraud Detection Using

Machine Learning
Suresh K Shirgave, Chetan J. Awati, Rashmi More, Sonam S. Patil

Abstract: In recent years credit card fraud has become one of the growing problem. A large financial loss has greatly affected individual person using
credit card and also the merchants and banks. Machine learning is considered as one of the most successful technique to identify the fraud. This paper
reviews different fraud detection techniques using machine learning and compare them using performance measure like accuracy, precision and
specificity. The paper also proposes a FDS which uses supervised Random Forest algorithm. With this proposed system the accuracy of detecting fraud
in credit card is increased. Further, the proposed system use learning to rank approach to rank the alert and also effectively addresses the problem
concept drift in fraud detection.

Index Terms: Concept drift, credit card fraud, Machine Learning, Random Forest
——————————  ——————————

1. INTRODUCTION Many learning algorithm have been presented for fraud

Credit card fraud is a major problem that involves payment detection in credit card which includes neural networks
card like credit card as illegal source of funds in transactions. [14][19][21][22], Logistic Regression [3], decision tree [4][15],
Fraud is an illegal way to obtain goods and funds. The goal Naive Bayes [6], Support Vector Machines [5], K-Nearest
of such illegal transaction might be to get products without Neighbors [6] and Random Forest [1][2]. This paper
paying or gain an unauthorized fund from an account. examines the performance of above algorithms based on
Identifying such fraud is a troublesome and may risk the their ability to classify whether the transaction was
business and business organizations. In the real world FDS authorized or fraudulent and then compares them. The
[1], investigator are not able to check all transactions. Here comparison is made using performance measure accuracy,
the Fraud Detection System monitors all the approved specificity and precision. The result showed that Random
transactions and alerts the most suspicious one. Investigator Forest algorithm showed better accuracy and precision than
verifies these alerts and provides FDS with feedback if the other techniques.
transaction was authorized or fraudulent. Verifying all the
alerts everyday is a time consuming and costly process. 2 RELATED WORK
Hence investigator is able to verify only few alerts each day. There are different supervised and unsupervised learning
The rest of the transactions remain unchecked until algorithms used for fraud detection in credit card. Some
customer identifies them and report them as a fraud. Also the important are described below. The author [1] has proposed
techniques used for fraud and the cardholder spending a paper where they have first explained the proper
behavior changes over time. This change in credit card performance measures which is used for fraud identification.
transaction is called as concept drift [1] [7]. Hence most of The authors have structured a novel learning technique that
the time it is difficult to identify the credit card fraud. Machine can solve concept drift, verification latency, and class
Learning is considered as one of the most successful imbalance issues. The paper also showed effect of above
technique for fraud identification. It uses classification and issues in true credit card transactions. Here in paper [2]
regression approach for recognizing fraud in credit card. The authors presented two types of classifier using random
machine learning algorithms are divided into two types, forests which are used to train the behavior features of
supervised [14][18] and unsupervised [16] learning transactions. The authors have compared the two random
algorithm. Supervised learning algorithm uses labeled forests and have analyzed their performance on fraud
transactions for training the classifier whereas unsupervised identification in credit card. In paper [3] authors presented a
learning algorithm uses peer group analysis [23] that groups FDS for credit card using Artificial Neural Network and
customers according to their profile and identifies fraud Logistic Regression. The system used to monitor each
based on customers spending behavior. transaction separately using classifier and then classifier
would generate score for each transaction and label this
_____________________________
 Dr. Suresh K. Shirgave is Associate Professor, DKTE Society’s
transaction as legal or illegal transaction. A decision tree
Textile and Engineering Institute, Ichalkaranji, India, E-mail: method was proposed in paper [4]. The method decreased
[email protected] overall misclassification costs and selected splitting property
 Chetan J. Awati is Assistant Professor, Department of Technology, at each node. The author also compared the decision tree
Shivaji University Kolhapur, India, E-mail: [email protected]
 Rashmi More is M. Tech. Student, Department of Technology, method for fraud identification with other models and proved
Shivaji University Kolhapur, India, E-mail: that this approach performs well using performance measure
[email protected] like accuracy and genuine positive rate. The author [5]
 Sonam S. Patil is Assistant Professor, Department of Information developed a FDS for credit card transaction using support
Technology, D Y Patil College of Engineering, Akurdi, Pune India, E-
mail: [email protected] vector machines and decision tree. This study built seven
alternative models that were created using support vector
machines and decision tree. The author also compared this
classifiers performance using performance measure

1217
IJSTR©2019
www.ijstr.org
 INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 8, ISSUE 10, OCTOBER 2019 ISSN 2277-8616

accuracy. The study also showed that as size of training Accuracy

dataset increases the number of fraud detected by SVM are 1.2 Precision
Specificity
less than fraud identified by decision tree method. Here in [6]
1
author presented fraud detection system using a Naive
Bayes K-Nearest Neighbors method. The main aim of 0.8
proposed system was to improve accuracy. Naive Bayes
0.6
Classifier predicts probabilities of fraud in transaction while
KNN classifier predicts how near the undefined sample data 0.4
is to kth training dataset. The author compared both this
0.2
classifier and showed that both work differently for given
dataset. Most of predictive model used for detecting fraud in 0
credit card transaction faces the issue of concept drift. The RF LR KNN SVM DT Navie
author [7] presented two FDS based on sliding window and Bayes
ensemble learning and showed that classifier need to be
trained separately using feedback and delayed samples. The
outcome of the two was than aggregated to improve the alert Fig. 1. Accuracy, precision and specificity performance of all
precision in FDS. Thus the author showed that to solve the classifier
issue of concept drift, the feedback and delayed samples are
to be handled separately.
From table 1 we can see that accuracy of random forest is
3 COMPARISON far better than the other learning algorithms. From fig. 1 we
can see that precision, accuracy and specificity of Random
Performance of all learning algorithms used for fraud
Forest is highest followed by Logistic regression, SVM,
detection in credit card transactions are compared in table 1
Decision Tree, Naive Byes and KNN. Hence the proposed
The comparison is based on their accuracy, precision and
system using random forest will show better accuracy for
specificity.
larger number of training data.
TABLE 1
4 PROPOSED SYSTEM
COMPARISON OF MACHINE LEARNING TECHNIQUES
Today modern society is using credit cards for variety on
Metrics reasons. Similarly fraud in credit card transactions has been
Classifiers growing in recent years. Each year, a huge amount of
Accuracy Precision Specificity financial losses are caused by the illegal credit card
transactions. Fraud may occur in variety of different forms
Random
0.962 0.997 0.987 and may be limited. Therefore there is need to solve the
Forest
issues of fraud detection in credit card. Additionally, with the
Logistic development of new technologies criminals finds new ways
0.947 0.996 0.979
Regression
to commit fraud. To overcome this problem the proposed
KNN 0.942 0.410 0.971
system for fraud detection in credit card transactions will be
designed using ML technique that will provide investigator a
small reliable fraud alerts.
SVM 0.938 0.782 0.984

Decision 4.1 Objectives

0.908 0.91 0.912
Tree The proposed system will achieve following main objectives:
 To train the model using feedbacks and delayed
Naive
0.937 0.505 0.9741 samples and sum up their likelihood to identify alert.
Bayes
 To implement machine learning technique to
address concept drift and class imbalance issue.
 To develop a learning to rank approach to increase
alert precision.
 To introduce performance measure those are
considered in real-world FDS.
We propose a Fraud Detection System (FDS), which mainly
focuses on data driven model and learning to rank method. It
also focuses on alert feedback interaction that checks the
way recent supervised samples are provided. Fig. 2 shows
the block diagram of proposed system.

1218
IJSTR©2019
www.ijstr.org
 INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 8, ISSUE 10, OCTOBER 2019 ISSN 2277-8616

will be alerted. Hence only limited number of alerted

transaction is reported to investigators.

4.2.4 Ranking of Alert

This module, rank each alert based on correctness of
security question. This security questions will be created
every time whenever the transaction is identified to be
suspicious. The alerts are ranked using likelihood. If it is
found that an alert has greater probability than other alerts
then it is added to a queue and location of fraudster is
tracked. This feature makes system user friendly and helps
to file complaint against fraud.

5 CONCLUSION
This paper has reviewed various machine learning algorithm
Fig. 2. Block Diagram of Proposed System detect fraud in credit card transaction. The performances of
all this techniques are examined based on accuracy,
precision and specificity metrics. We have selected
4.2 Modules
supervised learning technique Random Forest to classify the
The system is proposed to have the following modules along
alert as fraudulent or authorized. This classifier will be
with functional requirements.
trained using feedback and delayed supervised sample.
a. Data Preprocessing
Next it will aggregate each probability to detect alerts.
b. Scoring Rule
Further we proposed learning to rank approach where alert
c. Classification of Alerts
will be ranked based on priority. The suggested method will
d. Ranking of Alert
be able to solve the class imbalance and concept drift
e. Performance Analysis
problem. Future work will include applying semi-supervised
4.2.1 Data Preprocessing learning methods for classification of alert in FDS.
In this module selected data is formatted, cleaned and
sampled. The data preprocessing steps includes following: REFERENCES
a. Formatting: The data which is been selected may [1] Jalinus, N., Nabawi, R. A., & Mardin, A. (2017). The
not be in a suitable format. The data may be in a file Seven Steps of Project-Based Learning Model to
format and we may like it in relational database or Enhance Productive Competences of Vocational
Students. In 1st International Conference on Technology
vice versa.
and Vocational Teacher (ICTVT 2017). Atlantis Press.
b. Cleaning: Removal or fixing of missing data is called Advances in Social Science, Education and Humanities
as cleaning. The dataset may contain record which research (Vol. 102, pp. 251-256).
may be incomplete or it may have null values. Such [2] Andrea Dal Pozzolo, Giacomo Boracchi, Olivier Caelen,
records need to remove. Cesare Alippi and Gianluca Botempi, ‖Credit card Fraud
c. Sampling: As number of frauds in dataset is less Detection : A realistic Modeling and a Novel Learning
than overall transaction, class distribution is Strategy‖, IEEE Trans. on Neural Network and Learning
system,vol.29,No.8, August 2018.
unbalanced in credit card transaction. Hence [3] Shiyang Xuan,Guanjun Liu,Zhenchuan Li,Lutao
sampling method is used to solve this issue. Zheng,Shuo Wang, Jiang,‖Random Forest for credit card
fraud detection‖,Int.conf.on Networking,Sensing and
4.2.2 Scoring Rule control,2018.
Percentage of fraud in transaction is called as score. This [4] Y. Sahin , and Duman,E.,(2011) ―Detecting credit card
module assigns score by matching recent transaction pattern fraud by ANN and logistic regression.‖ In Innovations in
with the past transaction pattern of cardholder. If score is Intelligent Systems and Applications(INISTA),2011
greater then the transaction is considered as suspicious and international Symposium on (pp.315-319).IEEE
further proceeding is stopped. Otherwise it is moved to next [5] Y. Sahin, S. Bulkan, and E. Duman, ―A cost-sensitive
decision tree approach for fraud detection,‖ Expert Syst.
module.
Appl., vol. 40, no. 15,pp. 5916–5923, 2013
4.2.3 Classification of Alert [6] Sahin Y. and Duman E. (2011),‖Detecting Credit Crad
Fraud by Decision Trees and Support Vector Machines‖,
Here machine learning model will be used that will train and International Multi-Conference Of Engineers and
update the data based on feedback and delayed samples. Computer Scientists(IMECS 2011),Mar 16-18,Hong
Classifier will be trained separately using feedback and Kong,Vol.1,pp.1-6
delayed samples and their probabilities will be aggregated to [7] Sai Kiran, Jypti Guru, Rishabh Kumar, Naveen Kumar,
identify alerts. Transaction that will be having high probability Deepak Katariya,‖Credit card fraud detection using

1219
IJSTR©2019
www.ijstr.org
 INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 8, ISSUE 10, OCTOBER 2019 ISSN 2277-8616

Naïve Bayes model based and KNN classifier‖, Int. transaction neural network vol. 8, no. 4, pp. 827-834,
Journal of Adv. Research , Ideas and Innovations in Jul.1997.
Technology,vol.4,2018. [24] D. J. Weston, D. J. Hand, N. M. Adams, C. Whitrow, and
[8] A. Dal Pozzolo, G. Boracchi, O. Caelen, C. Alippi, and G. P. Juszczak,―Plastic card fraud detection using peer
Bontempi, ―Credit card fraud detection and concept-drift group analysis,‖ Adv. Data Anal.Classification, vol. 2, no.
adaptation with delayed supervised information,‖ in Proc. 1, pp. 45–62, 2008.
Int. Joint Conf. Neural Netw., 2015,pp. 1–8.
[9] A. C. Bahnsen, D. Aouada, and B. Ottersten, ―Example-
dependent cost-sensitive decision trees,‖ Expert Syst.
Appl., vol. 42, no. 19,pp. 6609–6619, 2015
[10] A. Dal Pozzolo, O. Caelen, and G. Bontempi, ―When is
undersampling effective in unbalanced classification
tasks?‖ in Machine Learning and Knowledge Discovery
in Databases. Cambridge, U.K.: Springer, 2015
[11] N. Mahmoudi and E. Duman, ―Detecting credit card
fraud by modified fisher discriminant analysis,‖ Expert
Syst. Appl., vol. 42, no. 5,pp. 2510–2516, 2015
[12] A. C. Bahnsen, D. Aouada, A. Stojanovic, and B.
Ottersten, ―Detecting credit card fraud using periodic
features,‖ in Proc. 14th Int. Conf. Mach. Learn. Appl.,
Dec. 2015, pp. 208–213.
[13] Anuruddha Thennakoon, Chee Bhagyani, Sasitha
Premadasa, Shalitha Mihiranga,‖Realtime Credit Card
Fraud Detection Using Machine Learning ,‖Int. Conf. on
Cloud Computing, Data Science & Engineering,2019.
[14] S. Wang, L.L. Minku, and X. Yao, ―Resampling-based
ensemble methods for online class imbalance learning.‖
Trans, Knowl., Data Eng., vol 27, no. 5, pp. 1356-1368,
May 2015.
[15] Jan may Kumar Behera, Suvasini Panigrahi, ―Credit
Card Fraud Detection: A Hybrid Approach using Fuzzy
Clustering and Neural Network‖, 2015 IEEE Second
International Conference on Advances in Computing and
Communication Engineering.
[16] M. Carminati, R. Caron, F. Maggi, I. Epifani, and S.
Zanero, BankSealer: ―A Decision Support System for
Online Banking Fraud Analysis and Investigation‖,
Berlin, Germany: Springer, 2014, pp. 380–394
[17] R. J. Bolton and D. J. Hand, ―Unsupervised profiling
methods for fraud detection,‖ in Credit Scoring Credit
Control VII. London, U.K.: Imperial College London,
2001, pp. 235–255
[18] R. Elwell and R. Polikar, ―Incremental learning of
concept drift in nonstationary environments,‖ Trans.
Neural Netw., vol. 22, no. 10,pp. 1517–1531, 2011.
[19] S. Bhattacharyya, S. Jha, K. Tharakunnel, and J. C.
Westland, ―Data mining for credit card fraud: A
comparative study,‖ Decision SupportSyst., vol. 50, no.
3, pp. 602–613, 2011.
[20] Tao Guo ,Gui-Yang Li ,‖Neural data mining for credit
card fraud detection‖,Int.Conf. on Machine Learning and
Cybernetics, Sept 2008
[21] J. Gao, B. Ding, W. Fan, J. Han, P.S. Yu, ―Classifying
data streams with skewed class distributions and
concept drifts‖, IEEE internet comput., vol.12, no. 6, pp.
37-49, Nov 2008
[22] E. Aleskerov, B. Freisleben, and B. Rao, ―CARDWATCH:
A neural network based database mining system for
credit card fraud detection,‖ in Proc. IEEE/IAFE
Computat. Intell. Financial Eng., Mar. 1997, pp. 220–
226.
[23] J.R. Dorronsoro, F. Ginel, C. Sgnchez and C.S. Cruz,
―Neural fraud detection in credit card operations‖, IEEE

1220
IJSTR©2019
View publication stats
www.ijstr.org