Detecting Fraudulent Insurance Claims Using Random Forests and Synthetic Minority Oversampling Technique

Abstract

There has been a significant amount of growth in the number of fraudulent activities by the policy-holders over the last couple of years. Deliberately deceiving the insurance providers by omitting facts and hiding details while claiming for insurance has led to significant loss of money and customer value. To keeps these risks under control; a proper framework is required for judiciously monitoring insurance fraud. In this paper, we demonstrate a novel approach for building a machine- learning based auto-insurance fraud detector which will predict fraudulent insurance claims from the dataset of over 15,420 car-claim records. The proposed model is built using synthetic minority oversampling technique (SMOTE) which removes the class imbalance-ness of the dataset. We use random forests classification method to classify the claim records. The data used in our experiment is taken from a publically available auto insurance datasets. The outcomes of our approach were compared with other existing models based on various performance metrics.