Artificial Intelligence–Driven Fraud Detection and Prevention in Community Banks and Credit Unions: A Systematic Review of Financial Security and Credit Unions Tomer-Trust Outcomes

This systematic review examines the applications, effectiveness, and governance concerns surrounding artificial intelligence (AI) fraud detection and prevention in community banks, credit unions, and other types of financial institutions. The search engines Scopus, Web of Science, IEEE Xplore, ACM Digital Library, ScienceDirect, and Google Scholar were used to conduct the search in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses 2020 guidelines. 834 records were identified, and then the duplicates were removed (N=22), leading to the screening of 812 records based on the title and abstract. The final 39 publications were analysed using a qualitative and descriptive approach after the exclusion of 59 publications from the 98 full publication texts assessed, for various reasons. Supervised machine learning, anomaly detection, deep learning, graph analytics, behavioural modelling, explainable AI (XAI), and combinations of these approaches in terms of the extent of successful transaction monitoring and detection, suspicious-pattern recognition, alert prioritisation, and/or detection speed. Real-time risk scoring facilitated better interventions, integrity, regulatory compliance, more effective cybersecurity controls and the utilisation of investigative resources. However, there were also issues with labelled data, class imbalance, data quality, legacy systems, implementation costs, lack of model transparency, algorithmic bias, use of AI models, risk of cyber insecurity, and vendor dependency. Accuracy and security, honesty, and fraud prevention - via AI - have been a major contributor to credit unions tomers' satisfaction, whilst systems that caused privacy issues have led to credit unions tomer dissatisfaction. The evidence strongly suggesting the participation of community banks and credit unions was less prevalent; generally, there were fewer bank- or credit union-specific findings. The review in conclusion argues that there is a need for that realised through a human-centric approach, with gradual deployment of models that can be explained and audited for fairness, privacy mechanisms, ongoing testing, training of employees' skills, and the presence of readily available credit unions tomer-redress mechanisms to guarantee that predictive abilities turn into institutional trust.