Machine Learning Based Method for Insurance Fraud Detection on Class Imbalance Datasets With Missing Values

Insurance fraud is a major challenge for the insurance industry, causing significant financial losses every year. Detecting fraudulent claims is difficult because fraud cases are rare compared to legitimate claims, resulting in highly imbalanced datasets. In addition, real-world insurance datasets often contain missing values and complex feature relationships, which further complicate fraud detection. This project proposes a machine learning-based approach using Super Learning (ensemble learning) and Explainable Artificial Intelligence (XAI) to improve fraud detection performance. The dataset used contains various insurance claim attributes such as policy details, incident information, and claim amounts. Data preprocessing techniques are applied to handle missing values and categorical variables, and class imbalance is addressed using SMOTE. Five machine learning algorithms are implemented and compared, including Logistic Regression, Decision Tree, Random Forest, Support Vector Machine (SVM), and XGBoost. These models are combined using a Super Learning framework to improve predictive accuracy. Explainable AI techniques such as SHAP and LIME are used to identify the most influential features contributing to fraud predictions. Experimental results show that the Super Learner model achieves 93.8% accuracy, outperforming individual algorithms while maintaining interpretability through XAI methods.