Graph Neural Network-Based Nationwide Healthcare Fraud Detection and Financial Risk Intelligence for Medicare and Medicaid Systems

Abstract

Healthcare fraud, waste, and abuse remain major financial and operational threats to Medicare and Medicaid systems, where fraudulent claims, provider collusion, upcoding, phantom billing, and abnormal referral patterns can generate substantial public expenditure losses. Traditional fraud detection methods, including rule-based systems, statistical models, and conventional machine learning, often analyze claims as isolated records and may fail to capture the complex relationships among providers, beneficiaries, procedures, diagnoses, pharmacies, facilities, and payment networks. This paper proposes a Graph Neural Network-based nationwide healthcare fraud detection and financial risk intelligence framework for Medicare and Medicaid systems. The proposed architecture models healthcare claims as a heterogeneous graph, enabling relational learning across multiple entities and supporting the identification of suspicious providers, anomalous claims, hidden fraud communities, and high-risk financial patterns. Building on graph anomaly detection, imbalanced learning, explainable artificial intelligence, and cost-sensitive fraud analytics, the framework integrates graph construction, GNN-based representation learning, fraud risk scoring, explainability, and audit-support dashboards. The paper contributes a scalable and policy-relevant model for improving payment integrity, strengthening public insurance oversight, prioritizing fraud investigations, and supporting data-driven financial risk intelligence across nationwide healthcare programs.