Abstract: Roadways, significant carriers of urban traffic, are essential to city safety improvement. Crash prediction models assist traffic administrators in identifying risk factors and estimate crash frequency, which play an essential role in traffic safety management. With crash occurrence and influencing factors change over time, however, the crash prediction models might not be suitable for current circumstances and even provided the wrong estimation for crash prediction. In order to explore the change of risk factors and crash frequency, this study conducted a longitudinal safety comparison of the urban roadway in Guangzhou, China. Utilizing the Bayesian negative binomial model framework, the relationships of crashes and safety influencing factors, such as road geometric characteristics, traffic operation characteristics, and road isolation facilities, have been accurately captured. Additionally, a two-stage Bayesian updating method was adopted to update the crash prediction model for 2020, based on informative prior information obtained from 2015. Modeling results indicated that updating an existing model is better than establishing a new model. Moreover, safety influencing factors had significant differences towards crashes longitudinally. The findings could be applied to long-term risk factors and hot spots identification, and more effective and well-targeted improvement measures can be implemented.
Xuesong Wang*, Chunting Nie, Zhicheng Dai. Roadway Crash Prediction Model Updating in Guangzhou, China. Transportation Research Board 101th Annual Meeting, Washington D.C., USA, 2022. 1.9-13.