Abstract : The integrity of the healthcare system is compromised by collusive fraud, where multiple individuals conspire to deceive health insurance providers. Despite the severity of this issue, current statistical and machine learning-based approaches struggle to identify fraudulent activities in health insurance claims, largely due to the similarity between fraudulent behaviors and legitimate medical visits, as well as the scarcity of labeled data. To enhance the accuracy of fraud detection, it is essential to incorporate domain expertise into the detection process. Through collaboration with health insurance audit experts, we have developed Fraud Auditor, a novel three-stage visual analytics framework designed to detect collusive fraud in health insurance. The approach begins with an interactive module that enables users to construct a co-visit network, providing a comprehensive representation of the relationships between patient visits. Next, we developed an enhanced community detection developed algorithm that incorporates fraud likelihood scores to identify high-risk clusters of suspicious activity. This is followed by a visual analytics interface that enables users to examine, compare, and validate suspicious patient behavior through customizable visualizations that accommodate various time granularities. To evaluate the efficacy of the current approach, we conducted a real-world case study in a healthcare setting, aimed at distinguishing actual fraudulent groups from false positives. The results, corroborated by expert feedback, demonstrate the effectiveness and usability of our methodology in supporting fraud detection and mitigation efforts.