A Generative AI-Enhanced Sociotechnical Framework (FRAIG) for Healthcare Fraud Detection: A Systematic Review and Evidence-Based Design

Abstract

Healthcare fraud detection presents significant challenges due to highly imbalanced datasets, evolving fraud patterns, and increasing regulatory requirements. While artificial intelligence (AI) approaches have demonstrated strong predictive performance, existing studies often lack methodological standardization, interpretability, and real-world applicability. This study presents a systematic review of 59 studies on AI-based healthcare fraud detection, identifying key methodological trends, performance characteristics, and critical limitations. The findings reveal a shift from traditional supervised learning approaches toward hybrid, deep learning, and graphbased models, alongside persistent challenges related to class imbalance, explainability, temporal validation, and deployment readiness. Based on these insights, we propose the FRAIG (Fraud Detection with Responsible AI and Generative Intelligence) framework, a novel sociotechnical architecture that integrates generative AI capabilities into healthcare fraud detection systems. FRAIG incorporates a generative augmentation layer for synthetic data generation and automated explanation, hybrid human-AI validation, interpretability-first model design, and regulatory compliance mechanisms aligned with healthcare standards. The framework is further supported by a practical use case and an evidence-based mapping of framework components to the literature, highlighting both well-established areas and critical research gaps. By bridging technical innovation with institutional and ethical considerations, FRAIG provides a comprehensive and adaptive approach to healthcare fraud detection. The study contributes to the literature by advancing sociotechnical systems theory through the integration of generative AI and by offering a scalable, regulation-aware framework for real-world deployment. Future research should focus on standardized evaluation protocols, empirical validation in operational settings, and the integration of privacypreserving and adaptive learning techniques.

Keywords: FRAIG, Medicare, Health Fraud, Systematic Review, Predictive Analysis