Analysis of driver behavior based on machine learning algorithm using vehicle driving data in racing track

Abstract

In this paper, we propose a driver behavior profiling method considering the several road shapes. Existing driver behavior profiling methods did not consider the road environment or only considered a specific driving environment. However, in order to profile driver behavior regardless of the situations that the driver is driving on, it needs to consider the general road environment such as road shape. To this end, we classify the several road types by applying the k-means clustering with dynamic time warping distance metric to the road curvature of the driving trajectory. Then, we profile driver behavior for each road type based on the driving data in each road type. For the evaluation, the proposed model is trained on one track and tested on the other track to verify whether it works well even in the novel road. The experimental results show that our model achieves 15% of improvement on average in driver identification over the existing method.