A Heterogeneity-Aware Car-Following Model: Based on the XGBoost Method

Abstract

With the rising popularity of the Advanced Driver Assistance System (ADAS), there is an increasing demand for more human-like car-following performance. In this paper, we consider the role of heterogeneity in car-following behavior within car-following modeling. We incorporate car-following heterogeneity factors into the model features. We employ the eXtreme Gradient Boosting (XGBoost) method to build the car-following model. The results show that our model achieves optimal performance with a mean squared error of 0.002181, surpassing the model that disregards heterogeneity factors. Furthermore, utilizing model importance analysis, we determined that the cumulative importance score of heterogeneity factors in the model is 0.7262. The results demonstrate the significant impact of heterogeneity factors on car-following behavior prediction and highlight the importance of incorporating heterogeneity factors into car-following models.