Performance Evaluation and Identification of Optimal Classifier for Credit Card Fraudulent Detection

Abstract

With the increase in the usage of the credit card by the people, the transaction done by credit cards has increased dramatically in the world. With this drastic increase in the usage of credit cards, the number of fraudulent also increases enormously and it is very difficult to identify the difference between a fraudulent transaction and normal transaction. American Express-issued credit card to 53.7 Million users, however, recorded Rs. 73,380 fraud in a year on average. The credit card fraudulent causes serious losses to the individual and the organization. The credit card issuing companies offer credit card fraud detection applications to the users and individuals for their safety. This paper focuses on the different algorithms used for credit card fraud detection and to find the optimal algorithm for classification of credit card fraud detection. It uses Logistic Regression, Linear Discriminant Analysis, K-Nearest Neighbors, Support Vector Machine, eXtreme Gradient Boosting, Random Forest and computes the accuracy, AUC-ROC values for all the classifiers.