AI & ML for Secure Payment Systems and Fraud Detection

Abstract

An increasing number of incidents concerning fraud through digital payments has become a major concern because of the growth in the number of payment transactions in online and mobile environments. It becomes increasingly difficult for conventional fraud detection systems based on rules to deal with complex and intelligent fraud attempts towards financial institutions. This paper intends to examine how artificial intelligence and machine learning assist in detecting fraud in a secure payment system. The quantitative research methodology was employed with respect to an artificial dataset of five thousand transactions. The models of Random Forest, XGBoost, and neural networks were created after class balancing using the SMOTE technique. The findings from the experiment show that both Random Forest and XGBoost were able to achieve more than 0.93 F1 and almost 0.999 AUC. The indicators that have been found to correlate the most with fraud detection were IP risk score and the number of transactions performed by the user. Confusion matrix analysis revealed the low levels of false positives and false negatives in the best-performing model. Latency analysis showed that all models had fulfilled the requirements for real-time transaction monitoring. The obtained conclusions may be used to confirm the effectiveness of applying machine learning models for fraud prevention in payment operations. Nevertheless, the use of simulated data and one-source indicators is a limitation of the research.