AI-Driven Evolution of Fraud Detection in Digital Banking

Abstract

The expansion of digital banking has greatly improved convenience for users but has also introduced new opportunities for financial fraud. Traditional rule-based fraud detection systems often fail to address the growing complexity and sophistication of modern cyber-attacks, highlighting the need for artificial intelligence (AI) in fraud detection strategies. This paper explores the application of AI, particularly machine learning models, in enhancing fraud detection within digital banking systems. Using the publicly available Banksim dataset, we evaluate six AI-based models—Random Forest, Logistic Regression, Naive Bayes, Decision Trees, Support Vector Machine (SVM), and K-Nearest Neighbor (KNN)—to assess their performance in identifying fraudulent transactions. The results show that Random Forest outperforms other models with the highest accuracy (96.5%) and AUC (0.97), followed closely by Logistic Regression. Our analysis demonstrates that AI-based models, especially ensemble learning techniques, provide a powerful, scalable solution for detecting fraud in digital banking. The findings underscore the critical need for financial institutions to adopt AI-driven approaches to bolster security and mitigate future fraud risks.