Enhancing Credit Fraud Detection in e-Payments through Predictive Analytics using Machine Learning: A Scoping Review

Abstract

The rapid global expansion of e-payment has created new opportunities for sophisticated, evolving financial fraud, leading to substantial financial losses and undermining or eroding customer trust, despite new security efforts. This scoping review examines how the application of predictive analytics, utilising machine learning and data analytics techniques, can enhance credit fraud detection and prevention in e-payments. The Joanna Briggs Institute protocols for scoping reviews were used to identify literature published between 2019 and 2025 from multiple academic databases, resulting in the selection of 30 relevant studies. Data mapping and analysis were applied to the selected studies, enabling insights into how financial institutions can leverage machine learning and data analytics to detect and prevent fraud more effectively across diverse digital payment environments. The review revealed various fraudulent methods, including complex transaction-like fraud, card-not-present fraud in online purchases, account takeover, identity theft, and high-frequency, small-scale fraud targeting vulnerable time windows, with emerging threats in DeFi platforms that pose unique detection challenges and high vulnerability. These findings demonstrate that fraud patterns are increasingly dynamic, adaptive, and designed to blend in with legitimate customer behaviour, making early detection more challenging. The study’s findings also highlight that advanced machine learning models, such as ensemble methods, deep learning, neural networks, anomaly detection algorithms, and data analytics methods, can significantly improve real-time fraud detection and outperform traditional rule-based approaches that rely on static thresholds, manual review, or historical assumptions. Despite these benefits, the review highlights persistent issues that limit practical implementation. These include data imbalance, high rates of false positives and false negatives, challenges with model transparency, privacy concerns, and integration challenges with legacy systems. Addressing these technical and operational challenges is therefore essential to enhancing the effectiveness of predictive analytics in safeguarding digital payment ecosystems and supporting a more trusted and resilient digital financial environment, while also informing regulatory compliance, operational risk management, and future research directions on fraud detection in e-payment systems.