Exposure based geographic analysis mode for estimating the expected pedestrian crash frequency in urban traffic zones; case study of Tehran

• Published in: Accident analysis and prevention 2022-04, Vol.168, p.106576-106576, Article 106576
• Main Authors: Almasi, Seyed Ahmad, Behnood, Hamid Reza
• Format: Article
• Language: English
• Subjects: Accidents, Traffic; Bayes Theorem; Environment Design; Exposure measures; Geographic expected crash frequency; Geographic weighted regression; Humans; Iran - epidemiology; Over-dispersion; Pedestrian crashes; Pedestrians; Reproducibility of Results
• ISSN: 0001-4575; 1879-2057
• DOI: 10.1016/j.aap.2022.106576

•The objective of this study is to estimate the expected geographical frequency of pedestrian crashes using the Empirical Bayesian (EB) approach using weighted geographical regression models for pedestrian crashes in Tehran.•In order to estimate the expected frequency of geographic crashes using pedestrian crash data in Tehran, four models of geographic weighted Poisson regression (GWPR), geographic weighted zero-inflated Poisson regression (GWZIPR), geographic weighted Negative Binomial regression (GWNBR) and the geographic weighted zero-inflated Negative Binomial regression (GWZINBR) have been used.•The overdispersion parameter is extracted from the four mentioned models and placed in the EB relation to obtain the geographic expected crash frequency.•The results showed that GWZIPR and GWZINBR models make more accurate predictions. These models had the lowest values of AIC and AICC, the lowest values of CV and the lowest values of MAD and RMSE. The Moran and VIF indices were also within acceptable limits. Predicting pedestrian crashes on urban roads is one of the most important issues related to urban traffic safety. Due to the lack of spatial correlation and instability in the crash data, the statistical reliability of Empirical Bayesian method in the combination of the observed and predicted crash frequency is questionable. In this study, an EB model has been developed to estimate the expected frequency of pedestrian crashes in urban areas using the over-dispersion parameter taking into account the spatial correlation of crash data. The objective of this study is to estimate the expected geographical frequency of pedestrian crashes using the Empirical Bayesian (EB) approach using weighted geographical regression models for pedestrian crashes in Tehran. For doing so, four models of geographic weighted Poisson regression (GWPR), geographic weighted zero-inflated Poisson regression (GWZIPR), geographic weighted Negative Binomial regression (GWNBR) and the geographic weighted zero-inflated Negative Binomial regression (GWZINBR) have been used. In this study, the areas analyzed for the development of the EB model based on pedestrian exposure variables include traffic analysis zones (TAZs). Finally, the EB model was extended to the Geographic Empirical Bayesian (Ge-EB) model. The results showed that GWZIPR and GWZINBR models make more accurate predictions. These models had the lowest values of Akaike Information Criterion (AIC), the lowest values of Cross Val