Medical insurance fraud, especially organized drug resale schemes, has become increasingly sophisticated, challenging traditional supervision methods. This paper presents an AI-powered legal supervision model that automatically detects fraudulent drug resale activities in medical insurance claims. Unlike rule-based approaches, our solution employs multi-dimensional behavioral analysis and adaptive clustering techniques to identify both individual anomalies and organized fraud networks. The proposed model follows a three-stage detection pipeline: (1) automated clue generation through feature aggregation across frequency, cost, and behavioral dimensions; (2) group behavior analysis using spatiotemporal patterns and medication similarity metrics; (3) risk stratification via FLASC clustering to dynamically determine suspicion thresholds. Key innovations include a data-driven threshold generation mechanism that eliminates expert bias and a cross-dimensional fraud pattern recognition system that connects individual outliers with group behaviors. Validated on real-world medical insurance data (8917 insurance cards, 1.1 million records), the model achieved 89% precision, 42% recall, and 87% accuracy in detecting high-risk fraud cases while uncovering previously unnoticed organized fraud rings. This research provides a scalable framework for intelligent healthcare fund supervision, with potential applications in other social security domains.
Indoor localization is a core technology for Internet of Things applications, but existing indoor positioning methods tend to suffer from poor positioning accuracy due to sensor noise and the short duration of signals collected during the cold-start phase. To address these issues, we propose CC-Loc, a cold-start multi-source collaborative localization framework that leverages geomagnetic signals and Bluetooth Low Energy (BLE) for infrastructure-free, high-precision positioning. Specifically, geomagnetic signals capture location-specific fingerprint features, while BLE RSSI models inter-device relative spatial relationships without precise distance conversion. However, the limited observation length at system startup inevitably introduces ambiguity in initial position estimation. To alleviate this issue, we design a cold start enhancement module that generates reliable predicted positions from short signal sequences. Building upon this initialization, we further enhance an encoder–decoder architecture with adjacency matrix–based spatial encoding and a Spatial Pyramid Pooling (SPP) layer to adapt to dynamic collaborative graphs and extract topological information. Additionally, we suppress BLE noise by a multi-task learning framework jointly optimizes the primary task, coordinate prediction and the auxiliary task, relative distance prediction. Experimental results show that our method achieves an error of only 2.93m in a trial site of 1750m^2, even when the length of the geomagnetic sequence at the initial startup phase is only one second. It effectively addresses cold-start challenges and adapts to dynamic multi-device collaboration, making it suitable for consumer-grade indoor localization applications.