An Explainable Hybrid Machine Learning Framew...

An Explainable Hybrid Machine Learning Framework for Detecting Upcoding and Duplicate Billing in Health Insurance Claims

Author Name : Akash N. Siddhpura, Dr. Pradyumansinh U. Jadeja

DOI: https://doi.org/10.56025/IJARESM.14102605

ABSTRACT

Health insurance fraud raises costs for insurers and, eventually, for patients. Two of its most common forms, upcoding and duplicate billing, are hard to catch because each fraudulent claim can look normal on its own. This paper presents an explainable hybrid framework that looks for these two patterns at the provider level. The framework builds behavioral features that compare each provider's charges, billing intensity, and repeated claims with those of its peers. K-Means clustering places providers into peer groups, and each provider's distance from its group center is used both to flag outliers with a three-sigma rule and as an input to a Random Forest classifier trained on SMOTE-ENN balanced data. Flagged providers receive short, plain-language explanations based on SHAP values and simple rules.