Artificial intelligence in health insurance fraud detection: a global and regional systematic review

Abstract

Health insurance fraud imposes substantial financial burdens on healthcare systems and reduces the efficiency of healthcare resource utilization worldwide. Traditional fraud detection approaches often struggle to analyse large and complex healthcare claims data, making artificial intelligence (AI) an increasingly important tool for identifying fraudulent activities. This study systematically reviews global and regional evidence on AI applications in health insurance fraud detection, with particular emphasis on methodological approaches, performance evaluation, and implementation challenges. This systematic review followed the PRISMA 2020 guidelines. Seven electronic databases were searched for studies published between 2016 and 2025. Titles and abstracts were screened using the Silvi AI platform, followed by eligibility assessment according to pre-defined inclusion and exclusion criteria. Data were extracted using a standardized framework, and methodological quality was assessed using the Drummond 10-item checklist. From 450 identified records, 33 studies met the eligibility criteria. The included studies demonstrated substantial heterogeneity in datasets, AI methodologies, fraud types, and performance evaluation metrics. Machine learning was the most frequently applied approach, followed by deep learning, graph-based learning, and hybrid AI frameworks. Although AI-based methods consistently improved fraud detection performance, challenges related to dataset imbalance, standardized reporting, interpretability, external validation, and economic evaluation remain significant barriers to widespread implementation. AI demonstrates considerable potential for improving health insurance fraud detection. However, future research should prioritize standardized evaluation frameworks, explainable AI, external validation across diverse healthcare systems, and economic assessment to support the responsible and scalable deployment of AI-enabled fraud detection systems.