The Use of Artificial Intelligence in Banking Fraud Detection: A Systematic Review Using the PRISMA Approach
The rise of digital banking has been accompanied by a corresponding rise in both the scale and the sophistication of financial fraud, thereby highlighting the structural limitations of traditional, rule-based detection systems. This study reviews the existing academic and industry literature on the use of artificial intelligence (AI) and machine learning (ML) in banking fraud detection using PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) framework. Following a clearly defined search and screening process, six full-text sources were used for full qualitative analysis. These include empirical survey research, hybrid statistical modelling, and prior systematic reviews. The study was also supplemented by a total of seventeen secondary references drawn from within those sources' own bibliographies. It was found that supervised learning, anomaly detection, deep learning, and ensemble techniques are the predominant techniques for fraud detection, with the highest improvement of ensemble and hybrid systems compared to single models. Empirical adoption studies rooted in the Technology Acceptance Model consistently identify technological knowledge, perceived usefulness, and organisational training — life more than algorithmic sophistication alone — as the most important predictors of long-term AI use. In addition, the most common barriers in the use of AI include shortage of skilled personnel, expensive computation, and evolving fraud tactics. Ethical and regulatory issues, including data privacy, algorithm bias, model transparency, and compliance with laws such as GDPR, PCI-DSS, and Anti-Money-Laundering law, emerge as a consistent and underdeveloped concern across the literature. The review summarizes that although AI represents substantial improvement in fraud-detection accuracy and speed when compared to rule-based systems, the existing remains geographically concentrated and methodologically descriptive, pointing to a need for more rigorous, cross-institutional, and cross-national research.