Abstract:
|
This study proposes a non-parametric Bayesian method to detect when the driving risk changes significantly for novice teenage drivers and clusters the drivers by their change-points and driving risk. Newly licensed teenage drivers are initially exposed to significantly higher accident risk and the risk patterns vary among subjects. We propose a Dirichlet Process mixture model allowing change-points to vary among drivers, while the change-points are assigned a Dirichlet Process prior. A Markov chain Monte Carlo algorithm is developed to sample from the posterior distributions. Automatic clustering is expected based on change-points without specifying the number of latent clusters. We apply the model to both simulated data and the Naturalist Teenage Driving Study data. Based on the Dirichlet Process mixture model, two clusters exist among the teenage drivers. The change-points of the two clusters are 69.12 and 98.12 hours. This research contributes to the knowledge of young drivers' driving behaviour and provides reference for parents, driver's education program, insurance companies, and vehicle regulations.
|