Predicting automobile insurance fraud using classical and machine learning models

Abstract

Insurance fraud claims have become a major problem in the insurance industry. Several investigations have been carried out to eliminate negative impacts on the insurance industry as this immoral act has caused the loss of billions of dollars. In this paper, a comparative study was carried out to assess the performance of various classification models, namely logistic regression, neural network (NN), support vector machine (SVM), tree augmented naïve Bayes (NB), decision tree (DT), random forest (RF) and AdaBoost with different model settings for predicting automobile insurance fraud claims. Results reveal that the tree augmented NB outperformed other models based on several performance metrics with accuracy (79.35%), sensitivity (44.70%), misclassification rate (20.65%), area under curve (0.81) and Gini (0.62). In addition, the result shows that the AdaBoost algorithm can improve the classification performance of the decision tree. These findings are useful for insurance professionals to identify potential insurance fraud claim cases.