Modeling Car-Following Behavior in Downtown Area based on Unsupervised Clustering and Variable Selection Method

In this research, an innovative framework that taking advantage of unsupervised clustering and variable selection method is proposed for the modeling of car-following behavior, suitable for incorporating explainable microscopic traffic models into understanding driver behavior. The proposed framework retains the advantages of both conventional and data-driven method. The experimental result presented in this paper shows that the unsupervised clustering method helps identify driver behaviors naturally in an intelligible way, while variable selection has shown a good property of identifying the true model of driving task while efficiently reducing model complexity. Especially, the proposed framework is demonstrated using real-world data collected from a sequence of instrumented install on a driving vehicle in Sakae, downtown area of Nagoya city, Japan. Gazis-Herman-Rothery (GHR) models, one of the most extensively used non-linear car-following models is calibrated against the same data and used as a reference benchmark.