Predicting credit card fraud using multipurpose classification based on evolutionary rules

Abstract

Due to advances in Internet technology, credit card transactions are increasing faster than ever before. This has led to a fraud problem that affects businesses, cooperating institutions, and government agencies. Credit card fraud occurs when a third party uses your credit card or credit account for an unauthorized transaction. Given the increased security of credit card transactions, fraudsters are developing new tricks or exploiting new vulnerabilities. Thanks to evolving technology, it is possible to analyze the use of maliciously obtained data by studying the time and cost associated with account switching transactions. In this study, we propose the ENORA and NSGA-II methods to create rule-based classifiers that can be easily interpreted using the credit card fraud diagnosis dataset created by analyzing the time intervals of individuals' credit card usage. In this experiment, credit card payments are classified as fraudulent or nonfraudulent based on several variables. The experiments were conducted in full training mode and 10-fold cross-validation mode. The performance of the algorithm was measured by accuracy, area under the receiver operating characteristic (ROC) curve, mean square error, proportion of true positives, proportion of false positives, precision, recall, F-measure, area under the curve (AUC), and Matthews correlation coefficient. The classification performance of the model using the correct classification ratio (ACC) is as follows: NSGA-II = 94.244%, ENORA = 93.236%. The performance of the other metrics is also discussed in detail in the results section. These results illustrate the significant predictive power of the proposed credit card theft models. After a thorough statistical analysis of our results, we found that the proposed strategy is capable of producing accurate and easy-to-understand categorization models.