Insurance Fraud Detection Using Machine Learning and XAI Algorithms

Insurance fraud poses a major challenge for the industry, raising expenses and premiums for insurers and policyholders alike. Deceptive claims raise operational expenses and cause elevated premiums, ultimately affecting the trustworthiness and effectiveness of the insurance sector. This study aims to investigate insurance fraud using 10 different machine learning models, namely Random Forest, Support Vector Machines, Gradient Boosting, K-nearest Neighbors, GaussianNaive Bayes, Extreme Gradient Boosting, Voting Classifier, AdaBoost, Decision Tree, and Logistic Regression. Among the models tested, the Voting Classifier model achieved the best performance, achieving both the F1-Score and the classification accuracy of 87%. Explainable Artificial Intelligence techniques, LIME and SHAP, are applied to explain the models' decisions. Both LIME and SHAP results show that incident_severity is the most significant feature in detecting insurance fraud.