A Data Mining Approach to Auto Insurance Fraud

Fraud is one of the most important problems that impose huge losses to insurance companies. There are many ways to do fraud detection; one of them is analyzing the historical data of detected frauds. Nowadays, using data mining methods to find the patterns of data is in common. Using these methods could be useful to identify spurious claims. Here, in addition to surveying the common methods of fraud detection in Auto insurance, three special methods, namely, Logistic Regression, Naive Bayes and Decision Tree, are applied to real data including 72 Auto insurance claims and their efficiencies are evaluated. As a result, Naive Bayes method has been chosen as the best identifier of real or fraud claims with 90.28% accuracy. Decision Tree and Logistic Regression, respectively stand in the next places, with 88.9% and 86.1% of accuracy.