KDE-OCSVM model using Kullback-Leibler divergence to detect anomalies in medical claims

Abstract

Detecting and preventing unusual claims behavior in medical insurance can reduce the drawdown of a group medical insurance pool. Taking the case of the urban employee basic medical insurance (UEBMI) in China, this paper develops a method to detect unusual medical claim patterns in the UEBMI. We collect public domain records involving gastric malignancy as the experimental sample. A method to preprocess the experimental sample for data learning is provided. We present a feature selection method, involving variance analysis and similarity analysis, to determine the core features. Next, we establish an extended one-class support vector machine (OCSVM) model, the kernel density estimation (KDE)-OCSVM, which exploits the Kullback-Leibler divergence and the KDE method to estimate the parameter v of the OCSVM model, to improve model performance. An experiment and two analyses are performed to validate the proposed method.