基于Cluster-GCN的医疗保险欺诈检测研究

Research on Detection of Medical Insurance Fraud Based on Cluster-GCN

摘要: 医疗保险欺诈是现阶段医疗保险机构面临的重大挑战，由于保险基金规模的扩张，欺诈导致的基金损失也在增加。随着大数据技术的发展，医疗保险欺诈检测技术在不断提高。本文针对个人的多条就诊记录与多个就诊医院，使用基于图聚类(社群发现)改进GCN的Cluster-GCN方法来搭建被保险人与医院、被保险人之间的关系网络，并设定行为信息更新与聚合的方式。检测结果表明，Cluster-GCN在医疗保险欺诈检测中有较好的分类性能，AUC达到0.86以上，总体识别的准确精度达到83.37%，反欺诈的识别率为77.25%。

Abstract: Medical Insurance fraud is a major challenge faced by medical insurance institutions at this stage. Due to the expansion of the insurance fund scale, the fund loss caused by fraud is also increasing. With the development of Big data technology, medical Insurance fraud detection technology is constantly improving. This article focuses on multiple individual medical records and multiple hospital visits and uses the improved Cluster-GCN method based on graph clustering (community discovery) to build a relationship network between insured individuals and hospitals, as well as between insured individuals. It also sets a way to update and aggregate behavioral information. The detection results show that Cluster-GCN has good classification performance in the detection of medical Insurance fraud, with an AUC of more than 0.86, overall recognition accuracy of 83.37%, and anti-fraud accuracy of 77.25%.