A Zero-Inflated Negative Binomial Crash Prediction Model for Freeway Bridge Sections

The frequent occurrence of crashes on freeway bridge sections has caused widespread concern. This paper takes the freeway bridge section as research unit and uses the homogeneity method to divide the section units of the E’dong Yangtze River Highway Bridge. Monthly traffic crash, traffic flow, and road geometric data in 2018 were used. A zero-inflated negative binomial (ZINB) regression model was developed to explore traffic safety influencing factors for freeway bridges. AIC criterion and BIC criterion were selected for the goodness of fit test. Relative error (Re), cumulative residual (CSR) and mean standard absolute residual (MSAD) were used for the accuracy test. The results showed that the ZINB model can accurately predict the number of freeway bridge crashes. This paper provides theoretical support and reference for improving the safety level of the freeway bridge operation.