BEAD-MI: Blending Ensemble Learning-based Anomaly Detection Model For Medical Insurance Data

Medical insurance fraud is a serious social problem, the effectively identification of abnormal medical insurance data can improve the security of medical insurance. The use of machine learning-based ensemble strategy can accurately detect abnormal data than a single model, but existing ensemble learning-based methods have the problems of high time consumption and low detection accuracy when dealing with large-scale medical insurance data. To solve this problem, this paper proposes a blending ensemble learning-based anomaly detection model called BEAD-MI for medical insurance data, it utilizes the advantages of six base models including Decision Tree, Random Forest, Extra Trees, CatBoost, KNN and XGBoost to increase the diversity and feature processing ability; In addition, the blending ensemble strategy is used to integrate the detection results of multiple single models to solve the shortcomings of existing methods such as high generalization error rate and poor generalization ability; We also replace the frequent used Logistic Regression in the blending model with Random Forest to achieve accurate and stable detection results. Extensive experimental results on three publicly available medical insurance datasets show that the proposed BEAD-MI model outperforms five state-of-the-art anomaly detection models in terms of accuracy, precision and F1-measure, and it also has better stability.