International Journal of Engineering Trends and Technology (IJETT) – Volume 8 Number 3- Feb 2014

Supervised Machine (SVM) Learning for Credit

Card Fraud Detection
Sitaram patel#1, Sunita Gond*2,
#1
M.Tech Scholar BUIT, Bhopal M .P. India
#2
Asst. prof., BUIT Bhopal M .P. India

Abstract—: The growth of e commerce increases the money

transaction via electronic network which is designed for hassle
free fast & easy money transaction. The facility involves greater
risk of misuse for fraud one of them is credit card fraud which
can happen by many types as by stolen card, by INTERNET
hackers who can hack your system & get important information
about your card or by information leakage during the
transaction. Several researchers have proposed their work for
credit card fraud detection by characterizing the user spending
profile. In this thesis we are proposing the SVM (Support Vector
Machine) based method with multiple kernel involvement which
also includes several fields of user profile instead of only Figure 1: Credit Card Blocks
spending profile. The simulation result shows improvement in
TP (true positive), TN (true negative) rate, & also decreases the II. CREDIT CARD FRAUD DETECTION
FP (false positive) & FN (false negative) rate. Growth in communication network, increased internet
speed, easy wireless connectivity & lack of time causes
Keywords— MATLAB, e-commerce, online banking, the people to buy through electronic network. Here are
classification support vector machine some statistics and projections of the Indian credit card
industry (http://hubpages.com/hub/Indian-Credit-card-
I. INTRODUCTION
Industry) to show importance of the topic.
The prediction of user behaviour in financial systems can be  India is currently the fastest growing Mobile
used in many situations. Predicting client migration, Market in the world and is also among the fastest
marketing or public relations can save a lot of money and growing credit card markets in the world.
other resources. One of the .most interesting fields of  India has a total approx.75 million cards under
prediction is the fraud of credit lines, especially credit card circulation (25 million credit and 50 million debit)
payments. For the high data traffic of 400,000 transactions per and a 30% year-on-year growth.
day, a reduction of 2.5% of fraud triggers a saving of one  With 87% of all transactions in plastic money
million dollars per year. happening through credit cards, debit cards in
India continue to be used largely for cash
Certainly, all transactions which deal with accounts of known withdrawals.
misuse are not authorised. Nevertheless, there are transactions  Though Visa, which accounts for 70% of the total
which are formally valid, but experienced people can tell that card industry is the market leader in India;
these transactions are probably misused, caused by stolen MasterCard is fast catching up.
cards or fake merchants. So, the task is to avoid a fraud by a  Every transaction involves payment of an
credit card transaction before it is known as “illegal”. With an interchange charge to MasterCard or Visa for
increasing number of transactions people can no longer settlement, which amounted to about $50 million
control all of them. As remedy, one may catch the experience during the year.
of the experts and put it into an expert system. This traditional
approach has the disadvantage that the expert’s knowledge,  Internal estimates of Barclaycard have pegged the
even when it can be extracted explicitly, changes rapidly with Indian market with potential to grow to at least
new kinds of organized attacks and patterns of credit card 55million credit cards by 2010-11.
fraud. In order to keep track with this, no predefined fraud
models as in but automatic learning algorithms are needed. III. CARD CLASSIFICATIONS
Qualified: The lowest retail processing rate category, In order
for a transaction to qualify for this category, the credit card
must be swiped though the terminal, the contents of the

ISSN: 2231-5381 http://www.ijettjournal.org Page 137

 International Journal of Engineering Trends and Technology (IJETT) – Volume 8 Number 3- Feb 2014

magnetic stripe transmitted, and an authorization must be is preferred because of its authenticity proven by time &
received. The transaction must also be settled or batched out simple programming. It also contains specific tool box for
of the terminal within 24 hours of the authorization. machine learning functions called bioinformatics which
Mid-Qualified: The lowest rate category for which key- contains complete set of function required by our program this
entered transactions can qualify. This category requires that reduces the programming effort & increase the precision. The
the billing address of the cardholder be verified with a match main function from SVM tool box is used to train a machine
of their zip code. Once again, an authorization number must by passing machine parameters. Stretcher of train machine
be received, and the transaction must be settled or batched out which can be used for data classification in future through
of the terminal within 24 hours of the authorization. SVM classify.
Non-Qualified: The highest rate category in the processing
environment. Generally speaking, this rate applies to key- V. RESULTS ANALYSIS
entered transactions where the address is not verified with a The results are simulated for three different Fraud
zip code match, or transactions that are not settled or batched probabilities Kernel Type: Linear, Quadratic and RBF from
out of the terminal within 24 hours of their authorization. 0.30 to 0.50 & changing the training data size from 30 to
Most business cards also fall into the category of pricing. 100.

IV. PROPOSED ALGORITHM AND IMPLEMENTATION

Firstly input the transaction probability for which or for what
number of synthetic data you want to generate. It will be taken
as the number of day you have done. Some transaction on
your account according to the input the number of days will
be selected randomly. Once we get the synthetic data generate.
Here we detail the proposed algorithm for classification of
Fraud Transactions.
 Read the given data. .
 Re-categorize the data into five groups as transaction
month, date, day, amount of transaction & difference
between successive transaction amounts.
 Make each transaction data as vector of five fields.
 Make two separate groups of data named True &
False transaction group (if false transaction data is
not available add randomly generate data in this
group).
 Select one of three kernels (Linear, Quadratic, and
RBF). This shows that the RBF kernel outperform to Linear &
 Train SVM. quadratic kernel in all fields of comparison it has
 Save the classifier. maximum accuracy up to 97%, maximum
 Read the current Transaction. TPR(99%),maximum TNR(98%) & maximum
 Repeat the process from step1 to step3 for current FPR(7%),maximum FNR(6%), it also behaves almost same
transaction data only. for all type of data set generated(having very low fraud
 Place the saved classifier & currently generated data & high fraud data).
vector in classifier.
 Take the generated decision from the classifier. VI. CONCLUSION & FUTURE WORK
We describe the techniques which are used in implementing Referring to results we can say that proposed algorithm
our model. The forms as well as the output screens are also with RBF kernel gives the better accuracy in comparison
shown to explain the working of our implemented algorithm. with the linear, Quadratic. In previous papers HMM model
Since there is no real data is available because of privacy was used. The SVM classifier scheme is a novel scheme used
maintained by banks. Hence for testing & implementation of by us. In this we have compared performance in three
our algorithm we have generated the data of true & false different kernels namely linear, Quadratic, RBF. As shown in
transaction using different mean & variance & then mixed the result our approach is better than the previous approaches,
them with different probability. We have used MATLAB for hence it can be used for automatic Credit card Fraud detection
the implementation of the algorithm because of its rich sets of with excellent accuracy & minimum false alarm.
mathematical functions and also supporting the inbuilt A. Limitations
functions for SVM.  None of the machine learning classifier algorithms
The complete simulation & comparison of all three kernels are evaluated was able to perform detection of user-
performed in MATLAB 7.5 environment. The MATLAB 7.5 to-root attack categories significantly.

ISSN: 2231-5381 http://www.ijettjournal.org Page 138

 International Journal of Engineering Trends and Technology (IJETT) – Volume 8 Number 3- Feb 2014

 No procedure was devised for kernel width

delimitation that is to obtain the best kernel function. [13] M. Syeda, Y.Q. Zhang, and Y. Pan,( 2002). “Parallel Granular
Networks for Fast Credit Card Fraud Detection,” Proc. IEEE Int’l
B. Future Work Conf. Fuzzy Systems, pp. 572-577,
 Enhance this model for dynamic improvements in
training of classifiers using different SVM models [14] S.J. Stolfo, D.W. Fan, W. Lee, A.L. Prodromidis, and P.K.
like incremental SVM, detrimental SVM and Chan(1997),“Credit Card Fraud Detection Using Meta-Learning:
Issues and Initial Results,” Proc. AAAI Workshop AI Methods in
evolutionary SVM etc. Fraud and Risk Management, pp. 83-90,.
 Explore continuous updating and adaption of
subsystems and combiner. [15] S.J. Stolfo, D.W. Fan, W. Lee, A. Prodromidis, and P.K. Chan,
 Extend adaptive classifier to finite mixture model (2000) “Cost-Based Modeling for Fraud and Intrusion Detection:
Results from the JAM Project,” Proc. DARPA Information
(more flexible), approximate logistic regression and Survivability Conf. and Exposition, vol. 2, pp. 130-144,.
RBF networks.
 More realistically handle the delayed fraud label.
REFERENCES

[1] Abhinav Srivastava, Amlan Kundu, Shamik Sural, Arun

Majumdar,( Jan.-Mar. 2008) “Credit Card Fraud Detection Using
Hidden Markov Model," IEEE Transactions on Dependable and
Secure Computing, vol. 5, no. 1, pp. 37-48,.

[2] Wen-Fang Yu, Na Wang,( 2009.)“Research on Credit Card Fraud

Detection Model Based on Distance Sum International Joint
Conference on Artificial Intelligence ," JCAI, pp.353-356,

[3] Sushmito Ghosh and Douglas L. Reilly Nestor,( 1994.) “Credit

Card Fraud Detection with a Neural-Network,” Inc. Proceedings of
the Twenty-Seventh Annual Hawaii International Conference on
System Sciences,pp.621-630 ,

[4] Aihua Shen, Rencheng Tong (2007), Yaochen Deng School of

Management, Graduate University of the Chinese Academy of
Sciences, Beijing, 100084, China, “Application of Classification
Models on Credit Card Fraud Detection”,pp.1-4,IEEE.

[5] Chun-Hua JU, Na Wang,( 2009 ) “ Credit Card Fraud Detection

Model Based on Similar Coefficient Sum,” First International
Workshop on Database Technology and Applications ,pp.295 -
298 , Database Technology and Applications, First International
Workshop on Wuhan, Hubei .

[6] V. N. Vapnik,( 1995.)” The Nature of Statistical Learning

Theory,” New York: Springer-Verlag,Second Edition, Technical
report,

[7] V. Kecman,( 2001)“Learning and Soft Computing: Support Vector

Machines, Neutral Networks and Fuzzy Logic Models,”Cambridge,
MA: MIT Press,Technical report,

[8] Research Article Bird Species Recognition Using Support Vector

Machines Hindawi(2007) Publishing Corporation EURASIP
Journal on Advances in Signal Processing Volume, Article ID
8637,8pages doi : 10.1155 / 2007 / 38637.
[9] “Global Consumer Attitude Towards On-Line Shopping,”(Mar.
2007).http://www2.acnielsen.com/reports/documents/2005_cc_onl
ine shopping.pdf,

[10] D.J. Hand, G. Blunt, M.G. Kelly, and N.M. Adams,( 2000.) “Data
Miningfor Fun and Profit,” Statistical Science, vol. 15, no. 2, pp.
111-131,

[11] “Statistics for General and On-Line Card Fraud,( Mar. 2007)
“http://www. epaynews.com/statistics/fraud.html,.
[12] S. Ghosh and D.L. Reilly(1994.), “Credit Card Fraud Detection
with a Neural-Network,” Proc. 27th Hawaii Int’l Conf. System
Sciences:Information Systems: Decision Support and Knowledge-Based
Systems,vol. 3, pp. 621-630,

ISSN: 2231-5381 http://www.ijettjournal.org Page 139