Medicare Fraud Detection Using WTBagging Algorithm

Medicare fraud detection is essential for the stable development of the health insurance industry. In this study, we use the Bagging algorithm to build a Medicare fraud detection model. The Gradient Boost Tree, XGBoost, CatBoost, and Random Forest models, are proven effective in past studies, and are used as the base models to construct the Medicare fraud detection model. We proposed the Bagging algorithm based on the weighted threshold method named WTBagging and made ten model combinations using Bagging and WTBagging algorithms. The data are cleaned and sampled to construct three datasets with different class distributions. The 5-fold cross-validation process was applied to the model training and repeated ten times, and the F1 value was the performance metric to evaluate the model combination. The results show that the model combinations of the WTBagging achieved the highest F1 values under all datasets.