Enhancing Medicare Fraud Detection through ML: Addressing Class Imbalance with SMOTE-ENN

Abstract

Medicare fraud poses serious challenges, leading to considerable financial losses and damaging the integrity of healthcare systems. Conventional approaches to fraud detection often fall short due to the complex and ever-changing tactics used by fraudsters. This project focuses on improving the detection of Medicare fraud by utilizing machine learning techniques, particularly addressing the problem of class imbalance where fraudulent claims are far fewer than legitimate ones. We are developing a classification system that can differentiate between fraudulent and non-fraudulent Medicare claims using several advanced ML algorithms. These include XGBoost, AdaBoost, LightGBM, Decision Tree, Logistic Regression, and Random Forest classifiers. To tackle the issue of class imbalance, we implement SMOTE-ENN, which helps to balance the dataset and enhances the performance of our models. Our experiments reveal that using SMOTE-ENN significantly boosts the detection rate of fraudulent claims. By evaluating the models on both the imbalanced and balanced datasets, we observe notable improvements in essential metrics such as accuracy, precision, recall, and F1-score. Overall, our findings suggest that integrating SMOTE-ENN with ensemble learning techniques offers a strong method for detecting Medicare fraud effectively.