Patient Cluster Divergence Based Healthcare Insurance Fraudster Detection

Abstract

Healthcare insurance frauds are causing millions of dollars in loss for public healthcare funds around the world. Healthcare fraud detection methods can help us to avoid the loss of medical healthcare insurance fund and to improve medical quality. The existing fraudster detection methods always consider people who violate normal behavior patterns as fraudsters. However, fraudsters can evade these monitors by camouflage, by adding normal behaviors so that they look “normal.” Our focus is to spot healthcare insurance patient fraudsters in the presence of camouflage. Although camouflage may hinder fraudster detection to some extent, we find that camouflage behaviors always sustain in a short period when the fraudster is conducting fraud. In other words, camouflage behaviors will not last long. Hence, if we can consider the cluster divergence of each patients' hospital admission graph during a long time, we can detect healthcare insurance fraudsters free of the interference of fraudsters' camouflage behaviors. In this paper, we propose the patient cluster divergence-based healthcare insurance fraudster detection (PCDHIFD), which can get rid of the disturbance of camouflage in fraud detection. Extensive experiment results show that our PCDHIFD outperforms the comparison approaches in terms of f-measure by over 15%.