Extreme gradient boosting machine learning algorithm for safe auto insurance operations

Vehicle safety is one of the most important research topics not only for vehicular industry but also for auto insurance companies. It is of great significance for them to avoid compensating damaged vehicles and dealing with an incredible number of correct and fraudulent claims. In fact, fraudulent claims present a huge and a costly problem for insurance companies and end up with big losses reaching over billions of Dollars yearly. These frauds also have social-economical consequences as their costs are defrayed by the policy holder through the increase of their premiums to cover the insurer loss. Several insurance companies are exploring innovative solutions not only to improve customers safety and driving experience but also to streamline fraud detection methods since traditional ones are complex, time-consuming, and usually lead to inaccurate results. In this paper, we develop an automated fraud detection approach for auto insurance companies based on extreme gradient boosting algorithm, aka XGBoost. The objective is to automatically detect fraudulent claims and classify them into different fraud types. To this end, data analysis techniques are used to clean, explore, and extract relevant features. The proposed framework aims to minimize human intervention, deliver alerts for risky claims, and reduce monetary losses in the auto insurance industry. The obtained results reveal a high performance gain achieved by XGBoost in detecting and classifying fraudulent claims compared to other machine learning algorithms.