Automobile Accident Claims Fraud Prediction Based on Machine Learning

Automobile accident fraud represents a significant challenge for the insurance industry, causing massive annual losses and undermining market health. Conventional fraud detection methods fall short against novel and complex fraud schemes. This study uses an Oracle database of 15,420 auto accident claims from an related company. During data preprocessing, we handled missing values, removed outliers, and applied SMOTE for data balancing. Then, we built five machine learning models: LR, SVM, KNN, RF, and XGBoost. The XGBoost model outperformed the others in fraud identification. This research offers an efficient auto accident fraud detection algorithm and a data-driven risk control solution for insurers. Future work could explore time - series features, online learning mechanisms, and privacy - protection strategies to enhance the adaptability and reliability of fraud detection systems.