Identifying Hotspots Based on Finite-mixture Zero-truncated Negative Binomial Model

To accurately identify hotspots and propose safety countermeasures to improve the level of traffic safety,this paper proposes a finite-mixture zero-truncated negative binomial (FMZTNB) model structure based on a previously developed zero-truncated negative binomial (ZTNB) model,considering the heterogeneity of zero-truncated traffic accident data.The R software package and Markov chain Monte Carlo (MCMC) method were used to estimate the parameters of the ZTNB and FMZTNB models,respectively.Further,Gelman-Rubin convergence statistics were used to verify the MCMC process.A comparison of nine sites with three different safety levels using the two models indicates that the proposed FMZTNB model is more accurate in predicting the number of crashes.This study further identified hotspots with different percentages of high risk using the crash rate,ZTNB,and FMZTNB models.The identification results were evaluated using three consistency tests:accident number,discriminant point,and rank consistency tests.The analyses results reveal that crash-rate-based hotspot identification is not reliable;ZTNB-based models provide significantly better results in identifying risky roadway segments.In general,the FMZTNB model outperformed the ZTNB model in ranking road segments. © 2022 Xi'an Highway University. All rights reserved.