Non Linear-Logistic Regression Analysis for AI-Driven Medicare Fraud Detection

The Medicare Fraud Detection (MFD) field is constantly evolving and faces essential challenges, particularly when addressing imbalanced data issues. MFD also contends with missing values, high dimensionality, and the risk of overfitting. This paper proposes a Non- Linear Logistic Regression Analysis (NL-LRA) technique, which performs with high efficiency and interpretability in Medicare fraud detection. The technique incorporates Penalized Logistic Tree Regression (PLTR), which uses information from decision trees to improve the performance of logistic regression. Initially, data is obtained from the Medicare fraud detection dataset. Pre-processing is conducted to address the missing and duplicate values. The Synthetic Minority Oversampling Technique (SMOTE) is applied to solve the issue of data imbalance. Feature selection is performed using the Linear Discriminant Analysis (LDA) technique to select relevant data. The NL-LRA is used for classification, accurately detecting instances of Medicare fraud based on the patterns and interactions among various features. Finally, the NL-LRA accurately classifies Medicare fraud detection when compared to the existing techniques namely, CatBoost and Auto encoder. The proposed method achieves a high accuracy of 0.9920, precision of 0.9450, recall of 0.9909, and f1-score 0.9801on the Medicare fraud detection dataset.