Artificial Intelligence and Machine Learning Approaches for Healthcare Fraud Detection: A Review, Case Study, and Framework

Abstract

Healthcare fraud imposes a significant financial burden on payers worldwide, yet most detection systems still rely on static rule sets that struggle to capture evolving fraud patterns. In recent years, Artificial Intelligence (AI) and Machine Learning (ML) have shown clear advantages by learning complex behaviors from claims, provider, and utilization data. This paper presents a structured review of 42 studies published between 2015 and 2025, examining supervised, unsupervised, and hybrid approaches for healthcare fraud detection. The review highlights consistent performance gains from ensemble methods but also identifies recurring gaps, including limited external validation, sparse use of privacy-preserving collaboration, and weak linkage between model outputs and operational return on investment (ROI). Unlike prior surveys, this work contributes in three distinct ways. First, it synthesizes evidence with attention to real-world constraints such as interpretability, drift, and investigative capacity. Second, it includes a small-scale case study using the Kaggle Healthcare Fraud dataset, where Random Forest, XGBoost, and Isolation Forest models were compared; ensemble methods achieved the best balance of accuracy, recall, and F1-score. Finally, it proposes a deployable reference framework that integrates preprocessing, hybrid models, explainability, and federated options for secure collaboration. Together, these contributions provide both a comprehensive overview and actionable guidance for building scalable, investigator-friendly fraud detection systems.

Ähnliche Arbeiten