Health insurance fraud, especially collusive fraud involving multiple parties, is a serious threat to the sustainability and efficiency of healthcare systems. Traditional fraud detection methods, such as statistical methods and machine learning models, often fail to accurately detect such fraud due to the high similarity between fraudulent and legitimate medical behaviors, as well as the lack of sufficient labeled data. Such limitations often lead to high false positive rates and increased manual effort for verification by audit experts.To address these challenges, in this work we present FraudAuditor, a novel visual analytics based approach to enhance the detection of collusive fraud in health insurance systems. The proposed approach combines expert knowledge and computational methods in a three-stage process. First, a co-visit network is constructed to capture the relationships between patients based on their medical visit patterns. Second, a better community detection algorithm is used to detect suspicious groups by analyzing the fraud likelihood with weighted connections. Finally, an interactive visualization interface allows auditors to explore, compare and validate suspicious behaviors with contextual information, such as time, disease, drug usage and reimbursement details.The system overcomes the limitations of existing approaches by combining data-driven analysis with human expertise, which leads to improved detection accuracy and reduced false positives. The experimental validation based on real-world case studies shows the effectiveness and usability of the proposed system to detect real fraud groups and filter out the non-fraudulent ones. It enhances the efficiency of decision making and offers a feasible solution for fighting against complex collusive fraud in health insurance.
Johar et al. (Fri,) studied this question.