Credit Card Fraud Detection Using Various Classification and Sampling Techniques: A Comparative Study

Abstract

With an ascent in the development of web-based business, the utilization of credit cards for internet shopping has expanded significantly. This, in turn, has brought about a great deal of credit card fakes. However, once in a while. Consequently, the execution of effective fraud detection frameworks has turned out to be fundamental for all banks to limit their misfortunes as far as credit card transactions are concerned. Numerous advanced systems have been created to monitor different credit card exchanges in literature. In this way, individuals have been attempting their best to identify the extortion in credit card exchanges as much as they can. Various machine learning techniques have been applied to predict whether a particular transaction is fraudulent or not. The biggest challenge with the techniques is the unavailability of the balanced dataset. Which is due to the nature of the transaction: the fraud transactions are too less when compared to genuine transactions. This work handles the challenge by balancing the dataset. Five machine learning techniques: Random forest, Naive Bayes, Support Vector Machine, K-Nearest Neighbor and Logistic regression were applied on the balanced dataset with different sampling techniques such as Oversampling, Undersampling, Both sampling, ROSE and SMOTE. The performance metric AUC – ROC suggests that logistic regression performs with an accuracy of 97.04 % and precision of 99.99%.