Credit Card Fraud Detection through Data Mining

Abstract– With the increasing fraudsters day to day, the fallacious transaction is rapidly growing thus making the frauds in this scenario a matter of high importance. Huge database patterns are identified by various data mining models assisting to strengthen and detect the credit card fraud. This research primarily focusses on “Credit Card Fraud Detection” analyzing two different methods used to detect fraudulent systems and implement the new technologies to the system to minimize frauds in credit card transactions

Keywords- Data Mining, Methods, Credit Card, Fraud Detection, Hybrid Technology, Neural Network, Support Vector System

Data Mining is filled with application growth opportunities and research which are reliable and usable from the data. Rapid development of e-commerce, usage of credit card has become well known mode for online and regular transactions, increasing the credit card fraud simultaneously. Fraud detection is a complicated problem as unwanted transactions are hidden in the authoriosed transaction. Due to security reasons and also to gain trust of user’s credit card fraud detection has now become important to companies.

Get Help With Your Essay

If you need assistance with writing your essay, our professional essay writing service is here to help!
Find out more about our Essay Writing Service

Neural Networks [1], Bayesian Network [2], Hidden Naïve Bayes Network [12], Dempster Shafer [10] are few methods to be justified in detecting the credit card fraudulent system. With the new technologies, it has now become easy for the companies and banks to detect the fraudulent system.

With the growing credit card fraud problem in the industry this literature review will help us to understand and detect the techniques involved in detecting the fraudulent system. We will be describing two different approaches – Neural Network and Support vector machine approach thus learning a new method to minimize the fraudulent system. This paper helps us to analyze data mining methods with respect to credit card fraud system.

A. Neural Network Approach

Neural network fraud detection method is primarily based on working of a human brain. Just as the human brain is capable of learning things from the previous experiences and uses the knowledge to decide things occurring in day to day problems the same strategy is used while detecting a credit card fraud with Neural Network system. Neural Network can reflect a small part of complexity and regulation Banks use this kind of network method to detect the credit card fraud. The moment a transaction takes place there are a set of attributes attached to it characterizing the account holder, the amount and the merchant.

Considering an example, for the Mellow Bank Fraud Detection Feasibility Study a particular archived amount of data was used for model development as the authorized data wasn’t easily available due to security reasons. P-RCE [3] neural network technique is used. P-RCE is used for pattern recognition as it helps to describe what exactly the human brain is thinking about. P-RCE has a single cell layer which outputs a numeric response called as “Fraud score”. The lower the threshold the more no. of credit card fraud is detected. Higher the credit card detection threshold less no. of fraud is detected. With 2000000 transactions of Mellon Bank’s data from Oct-Nov 1991, nearly measuring 50 accounts per day 40% of the fraudulent transaction was observed but prior to use of P-RCE method in Mellon Bank’s feasibility study the result came out to be 1 fraudulent transaction per week on reviewing 750 account per day. The improvement in the fraud detection performance was undeniably considerable. The pattern recognition method can actually help the banks to reduce 20-40% (in total) credit card fraud losses.

B. Support Virtual Machine Approach

An online transaction has four entities: credit card holder, credit card, the seller and the buyer. There is an independent history of transactions with each set of entities in the data set. Each entity keeps almost a consistent behavior pattern in authorized transactions. The risk of a transaction can be estimated by calculating the inconsistency of a transaction from the history of authorized transactions. Let us define Xe(t) – {a1, a2, …, an) as the transactional set in the entity’s history is equivalent to f(t) beingF(t) – {seller, buyer, holder, card) – set of entities. We calculate the score of transactions, by defining l = (l1 …, l19) [3] with ‘c’ as value of a feature, as SC(l,e,t) = count(c,Xe(t)) The idea to calculate how near a transaction is related to past authorized transactions therefore we consider only those transactions that are proved legitimate from Xe(t).

To train and test a classifier (weight and score) we have a vector in conjunction with the classification: SVM (Support Vector Machines). They are essentially supervised learning models used for analyzing data and recognition of patterns. Using the traditional methods, raw data was classified with SVM to check the impact of general weight in the outcomes and reduce SVM complexity as well. SVM classifier was used in classifying the transactions as fraud or authorized resulting in 40-50% in most months with false alarm rate 10-12%.

The online credit card fraud system can be detected and improvised with the Big Data Technologies framework. The main aim is to achieve the goal of fusing various detection methods to enhance the accuracy. A workflow was proposed [13] containing common designs of fraudulent system thus making it easier to integrate identification of fraud system.

In recent years, big data platforms were released to process and operate data including MapReduce and Apache Hadoop frameworks (open source for MapReduce). Two components are primarily considered: spouts and bolts. The source of streams is referred to as spout which reads and sends tuples from external source into topology. The data processing is done by Bolt. With the reference, the paper proposes a hybrid structure with Big data efficient in solving challenges related to performance and integration. The basic workflow is defined [13] in the figure. QF- Quick filter, DSA – Demper Shafer Adder (combining different fraud scores and generating a merged result), EF- Explicit filter.

This workflow is designed by combining different algorithms together for a higher accuracy like two DSA’s are combined so that we can aggregate their fraud score to get better accuracy; 2) Another aspect can be considered by combining supervised & unsupervised fraud approaches to examine a good cover of types of fraud; 3) The filters used: QF – detects only the behavior of card holder whereas EF detects the historical data in the whole model. So, to achieve faster filter we can combine QF and EF as well as to get better efficiency the combination is good to go.

In this paper, we have reviewed two data mining detection methods of credit card fraud. The research papers which aren’t considered here might have comprehensive methods to research and implementations of new detection techniques. This research paper describes:

The Neural Network can be implemented in banks to reduce the credit card fraudulent system with it P-RCE algorithm. 2) Support Vector machine can detect the frauds in ecommerce system real time but isn’t much reliable for complex frauds. 3) Hybrid technology framework, the workflow is essential to detect frauds in offline system as the method but can improve accuracy, performance and efficiency.

To develop a credit card fraud system, the neural network method is best suited according to my understanding as it is efficient, accurate and cost effective thus implemented in Mellon Bank. Neural Network method has some failures as well but gradually with new technology it can overcome. But to develop a strong fraud detection system using credit card we need to combine few more complex detecting methods.

References

1. [1] S. Ghosh and D. L. Reilly, “Credit card fraud detection with a neural network, “in System Sciences, 1994. Proceedings of the Twenty-Seventh Hawaii International Conference on, vol. 3, Jan 1994, pp. 621-630.
2. [2] G. F. Cooper and E. Herskovits. “A Bayesian Method for the Induction of Probabilistic Networks from Data”. Machine Learning, 9(4):309-347, 1992.
3. Santiago, Gabriel Preti, Adriano Pereira, and Roberto Hirata Jr. “A modeling approach for credit card fraud detection in electronic payment services.” In Proceedings of the 30th Annual ACM Symposium on Applied Computing, pp. 2328-2331. ACM, 2015.
4. Online Credit Card Fraud Detection: A Hybrid Framework with Big Data Technologies You Daiâˆ-, Jin Yanâˆ-, Xiaoxin Tangâˆ-, Han Zhao† and Minyi Guoâˆ- âˆ-Department of Computer Science and Engineering, Shanghai Jiao Tong University, China †School of Computer Science & Technology, Huazhong University of Science and Technology, China
5. Philip K. Chan, Wei Fan, Andreas 1. Prodromidir, and Salvotore 1. Stalfo, “Distributed Data Mining in Credit Card Fraud Detection” 2016 IEEE TrustCom/BigDataSE/ISPA
6. H. Michael Chung & Fredric C. Gey “Data Mining, Knowledge Discovery, and Information Retrieval”, Proceedings of the 34th Hawaii International Conference on System Sciences – 2001
7. Agrawal, Ayushi, Shiv Kumar, and Amit Kumar Mishra. “Implementation of Novel Approach for Credit Card Fraud Detection.” In Computing for Sustainable Global Development (INDIACom), 2015 2nd International Conference on, pp. 1-4. IEEE, 2015.
8. [2008] Abhinav Srivastava, Amlan Kundu, Shamik Sural and Arun K. Majumdar, “CreditCard Fraud Detection Using Hidden Markov Model” IEEE, Transactions on Dependable and Secure Computing, Vol. 5, No 1., January-March
9. D. L. Reilly and L. N. Cooper, “An overview ofneural networks: early models to real worldsystems,” in An Introduction to Neural and Electronic Networks, ed. S. F. Zometzer, J. L. Davis and C. Lau, 227-248, Academic Press, (1990).
10. S. Panigrahi, A. Kundu, S. Sural, and A. Majumdar, “Credit card fraud detection: A fusion approach using dempstershafer theory and Bayesian learning,” Information Fusion, vol. 10, no. 4, pp. 354 – 363, 2009.
11. Z. D. Zhao and M. s. Shang, “User-based collaborative-filtering recommendation algorithms on hadoop,” in Knowledge Discovery and Data Mining, 2010. WKDD ’10. Third International Conference on, Jan 2010,pp. 478-481.

[12] Jiang, Liangxiao, Harry Zhang, and Zhihua Cai. “A novel Bayes model: Hidden naive Bayes.” IEEE Transactions on knowledge and data engineering 21, no. 10 (2009): 1361-1371.

[13] Dai, You, Jin Yan, Xiaoxin Tang, Han Zhao, and Minyi Guo. “Online Credit Card Fraud Detection: A Hybrid Framework with Big Data Technologies.” In Trustcom/BigDataSE/I​ SPA, 2016 IEEE, pp. 1644-1651.IEEE, 2016.