Prediction of automobile insurance fraud claims using machine learning

Abstract

Automobile insurance fraud is a significant issue for insurance firms, causing financial losses and higher premiums for policyholders. This study aims to create a predictive model for accurately identifying potential vehicle insurance fraud claims. Understanding fraud detection processes and operationalizing information communication technology is crucial for implementing corrective actions, but personally reviewing insurance claims is time-consuming and costly. This study explored machine learning algorithms to detect fraudulent vehicle insurance claims. The research evaluated AdaBoost, XGboostNB, SVM, LR, DT, ANN, and RF. AdaBoost and XGBoost classifiers outperformed other models with 84.5% classification accuracy, while LR classifiers performed poorly with balanced and unbalanced data. The ANN classifier performed better with unbalanced data. Performance evaluation metrics such as accuracy, precision, recall, and F1 score are utilized to assess the effectiveness of the models. The results demonstrate the effectiveness of machine learning in distinguishing between genuine and fraudulent claims, providing insurance companies with a powerful tool to proactively combat fraud and improve their overall risk management strategies. The findings of this research contribute to the insurance industry’s efforts to enhance fraud detection systems, reduce financial losses, and offer more competitive insurance premiums to honest policyholders.