Automobile Insurance Fraud Detection Based on PSO-XGBoost Model and Interpretable Machine Learning Method

Automobile insurance fraud has become a critical concern for the insurance industry, posing significant threats to socio-economic stability and commercial interests. To tackle these challenges, this paper proposes a PSO-XGBoost fraud detection framework and uses explainable artificial intelligence to interpret the predictions. The framework combines an XGBoost classifier with the particle swarm optimization algorithm and is validated through a comparative evaluation against other models. Traditional methods, including SVM, Naive Bayes, Logistic Regression, and BP Neural Network, demonstrate moderate accuracy, ranging from 54.1% to 68.6%, while more advanced models like Random Forest reach up to 78.4%. Compared to the standard XGBoost, the PSO-optimized model achieves 3% superior accuracy, achieving an impressive 95% success rate. Moreover, SHAP is used to extract and visually depict the contribution of each feature to the model's predictions. It turns out that the policyholder's claim amount is the most significant factor in detecting automobile insurance fraud, with other factors such as vehicle type, responsible party, and the insurer's age also considerably influencing the prediction performance. This paper therefore proves that combining the PSO-XGBoost model with SHAP approach can substantially improve the early warning and prevention of automobile insurance fraud.