Developing an Interpretable Driver Risk Assessment Model to Increase Driver Awareness Using In-Vehicle Records

Abstract

Providing feedback to drivers on their risky driving behaviors is an important method to improve drivers’ awareness in reducing future accidents. However, it is hard to identify risk-prone behaviors and explain them to drivers. In the present study, we used driving log from 103370 electric vehicles equipped with L2-assisted driving functions. We used 28 explainable features to establish a binary classification model of accidents and eight features can be used to establish an acceptable model. Further, we developed an easy-to-understand safety score formula using these eight features. Through this accurate and transparent feedback, we may improve drivers’ safety awareness without undermining their trust in the L2 and higher level automated vehicles. This will not only reduce accidents but enable them to adapt to the development of automated driving technology in a smoother manner.