Insurance fraud detection: Evidence from artificial intelligence and machine learning

Abstract

This study proposes a framework for fraud detection in the auto insurance industry by using predictive models. The feature selection is performed utilizing a publicly available car insurance dataset and uncovers the most influential feature through Boruta algorithm. Three predictive models (logistic regression, support vector machine, and naïve Bayes) are applied for developing a fraud detection mechanism. Six metrics are computed from the confusion matrix to assess the performance of the predictive model. The results reveal that the support vector machine outperforms in terms of accuracy, and the logistic regression achieves the highest f-measure score. Each influential feature's ranking is performed, and it is revealed that the fault, base policy, and age of the policyholder are the most influential features. The findings of this study are beneficial for fraud detection in the auto insurance industry. Additionally, the underlying framework holds a functionality for real-time problem-solving in the auto insurance industry.