A Novel Combination Approach of SHAP and Machine Learning with Explainable AI for Predicting and Explaining Vehicle Insurance Fraud
Аннотация
The detection of vehicle insurance fraud remains a key concern since high-performance artificial intelligence (AI) models cannot be completely trusted, while rule-based methods are too rigid to encompass shifting fraud tactics. Machine learning significantly enhances detection but is hindered by its black-box nature in highly regulated fields requiring clear and explainable decision-making. To bridge this gap, this study proposes an explainable AI (XAI) model for fraud detection in vehicle insurance claims using random forest and Shapley additive explanations (SHAP). SHAP provides post-hoc interpretability, allowing insurers to identify feature importance and fraud-related feature contributions in insurance claims. The model is evaluated on an insurance dataset comprising 28,836 vehicle insurance claims. The optimized model achieved an accuracy of 0.90, precision of 0.82, F1-score of 0.80, and an AUC-ROC of 0.92 on the independent test set. In addition to its technological contributions, the proposed model aligns the proposed fraud-detection framework with selected ISO 31000:2018 risk management principles to support transparent review, accountable decision-making, and systematic risk assessment. By incorporating XAI into fraud detection, the research provides interpretable information, supporting risk assessment and fraud-review processes while highlighting the broader applicability of explainable machine learning models in high-risk financial sectors.