Automobile Insurance Claims Auditing: A Comprehensive Survey on Handling Awry Datasets

Abstract

Fraud is a very costly criminal activity. Insurance companies face the very challenging task of identifying and preventing fraudulent claims. Just like any big problem in recent years, Machine Learning has been heavily applied to fraud detection in both a supervised and non-supervised manner. But, usually supervised models do not perform well in the presence of awry, asymmetrical Datasets. This paper presents a novel approach for auditing claims in automobile insurance. Our data pipeline consists of preprocessing, feature selection, data balancing, and classification. This robust fraud detection model, built upon existing fraud detection research, gives very promising results compared to state of the art in the industry.