AN EXPLAINABLE AI FRAMEWORK FOR AUTOMATED MEDICARE AND MEDICAID CLAIMS VALIDATION, ANOMALY DETECTION, AND IMPROPER PAYMENT RISK ASSESSMENT

Objective: This research aims to explore the patterns of healthcare claims, provider behaviour, and possible signs of payment risk using an explainable artificial intelligence approach that leverages the exploratory data analysis method. Method: Secondary data, including beneficiary, provider, inpatient, and outpatient claim data, were analyzed to look at distribution, relationships, and provider-level differences. Various descriptive statistical methods, comparative analyses, correlation assessment, and visual analytics were utilized to investigate reimbursement activity, claim frequency, beneficiary characteristics, and indicators of fraud. Results: The results show that providers with fraud indicators are a small fraction of the data set, but that there are significant differences in billing/reimbursing practices between this group and the rest of the providers. While the demographic variables show less correlation regarding payment patterns, there are meaningful correlations identified between the claim volume, reimbursement measures, and fraud indicators. Integrated anomaly analysis also identifies provider profiles that have unique claim and reimbursement patterns that could indicate that the provider is potentially underutilizing the program and could benefit from further administrative review. The study shows that explainable visual analytics can enhance the understanding of atypical billing patterns, potential claims validation issues, and provider-level payment risk characteristics, while not necessarily making fraud decisions. Novelty: The research highlights the importance of transparency, interpretability, and responsible handling of health care data, providing an analytical tool that can assist healthcare administrators, auditors, and researchers in analyzing intricate health care payment data and pinpointing areas for additional study.