Freeways play an important role in transportation system. But crashes occur frequently on freeways. Crash prediction models for freeway can evaluate road safety status and analyze safety influencing factors. Based on 9987 crashes occurred on Ningbo-Taizhou-Wenzhou Freeway from 2016 to 2019, this paper develops a Random Effect Negative Binomial (RENB) model, which introduces a random effect term into a Negative Binomial (NB) model to explain time correlation of the crash data among different years. In this paper, the NB model and the RENB model are compared in terms of goodness of fit and prediction accuracy. Results show that factors influencing crashes are annual average daily traffic volume, length of research unit, number of lanes and road alignment; the goodness of fit of the NB model and the RENB model is consistent, while the NB model has better prediction accuracy.

Crash prediction model; time correlation; NB model; RENB model;