A Bayesian Gaussian Mixture Model for Probabilistic Modeling of Car-Following Behaviors
Xiaoxu Chen, Chengyuan Zhang, Zhanhong Cheng, Yuang Hou, Lijun Sun
Abstract
Car-following models are essential for microscopic traffic simulation. While conventional models rely on parsimonious formulas with simplified assumptions, recent studies have focused on developing data-driven models with the help of high-resolution trajectory data. This paper presents a data-driven model based on a Bayesian Gaussian mixture model (GMM) for probabilistic forecasting of human car-following behaviors. By incorporating past and future information, our model captures the temporal dynamics of human car-following behaviors, providing accurate predictions of the following vehicle’s behavior and quantifying the forecast uncertainty. We demonstrate the interpretability of the Bayesian GMM in modeling car-following behaviors, providing valuable insights into the heterogeneity and uncertainty of driver behaviors. Additionally, we show that the proposed model can make probabilistic multi-vehicle simulations that reproduce natural traffic phenomena. Our results suggest that the proposed Bayesian GMM is a promising approach for modeling and forecasting car-following behaviors in various driving scenarios, contributing to the development of safer and more efficient transportation systems.