Evaluation and Interpretation of Driving Risks: Automobile Claim Frequency Modeling with Telematics Data

Abstract

With the development of vehicle telematics and data mining technology, usage-based insurance (UBI) has aroused widespread interest from both academia and industry. The extensive new driving behavior features make it possible to further understand the risks of insured vehicles, but pose challenges in the identification and interpretation of important ratemaking factors. This study therefore analyzes telematics data to predict UBI claim frequency. Specifically, based on 75 features theoretically related to driving risk, we employ both Poisson regression and several machine learning models to model claim frequency. After selecting the best model, we analyze feature importance, feature effects, and the contribution of each feature to the prediction from an actuarial perspective and illustrate the specific impact of driving behavior features through some visualization tools. The methodology and conclusions of this study not only provide a more accurate prediction, but also improve the practical value of machine learning in insurance risk assessment. This empirical study, based on the telematics data of policyholders in mainland China, shows that XGBoost greatly outperforms the traditional models and detects the complex relationships in the nonlinear and interactive effects of various features fully. In addition, according to the contribution of each feature, we develop a customer segmentation plan to distinguish high-risk drivers, which can serve as a guideline for insurance companies to achieve ex-ante risk management.