COMPARATIVE ANALYSIS OF CLASSIFICATION TECHNIQUES FOR CREDIT CARD FRAUD DETECTION

Abstract

Nowadays, in the global computing environment, online payments are a necessary evil as it makes payment conveniently easier and can be done via an ample of available options like a Credit card, Debit Card, Net Banking, PayPal, Paytm available to make payments easier. The most common mode of payment used in online shopping is Credit Card as it is easier for the customers to directly transfer money from one account to another; without the withdrawal of cash at any point. However, this easy payment mode has opened up paths for multiple frauds which involve theft or illegal tampering of data of the credit card owner. Thus, with the increasing number of fraud cases and losses, it is important to find the best solution to detect credit card fraud as well as minimize the number of frauds in online systems. With the analysis of different sets of research performed on the given problem statement, we have concluded that the issue requires a substantial amount of predictions and application of machine learning to find the accuracy score of those commonly used algorithms to predict which of these three state-of-art-algorithms - Naive Bayes, Logistic Regression and K Neighbours, is best suitable to carry out the research in this area. In order to support our findings, we apply two different approaches i.e. with sampling and without sampling on these algorithms against the same dataset. We claim on the basis of our results that K Neighbours outperformed all in both the approaches and is more suitable to carry forward the fraud detection research using machine learning. The analysis will be useful for those working to derive anti-fraud strategies to predict the fraud patterns and reduce the risk during hefty transactions.