Research on UBI Claims Risk Probability Models Based on Machine Learning

In recent years, with the development of big data and the popularization of the Internet of vehicles, driving behaviors can be effectively recorded and analyzed, and applied in various automotive post-market fields, including insurance pricing, model risk classification, and so on. With the reform of commercial vehicle rates, more and more insurance companies have the power to set their price. Compared with the traditional influencing factors, driving behavior has a more significant impact on the risk probability. Therefore, it is necessary to adjust the premium based on driving behavior. Based on the collected data of 100 driving behaviors and accident records, this paper uses the traditional Logistic Regression model and three popular machine learning algorithms: KNN algorithm, Random Forest algorithm, and XGBoost model to compare their prediction results on risk probability. By comparison, it is found that XGBoost model has higher prediction accuracy. Therefore, this algorithm is used to rank the importance of the influencing factors to the risk probability, to provide a basis for insurance companies to formulate subsequent premiums, and also, it can improve the driving behavior of drivers, and reduce the traffic accident rate.