Development of Planning-Level Transportation Safety Models using Full Bayesian Semiparametric Additive Techniques

Abstract

Recently, several attempts have been made to develop collision prediction models in which spatial dependency is considered. These models recognize the local nature of spatial data by relaxing the regression analysis assumption that the error terms for each observation are independent. The primary objective of this study is to investigate an alternative technique for capturing the spatial variations in the relationship between the number of zonal collisions and potential transportation planning predictors. Spatial relationships are incorporated into the full Bayesian semiparametric additive modeling framework through the covariance of the error terms. The secondary objective of this research study is to build on knowledge of comparing the accuracy of full Bayesian models to that of generalized linear and geographically weighted Poisson regression models. The spatial covariates from the full Bayesian semiparametric additive model indicate that collision frequencies in traffic analysis zones are spatially correlated. The results of accuracy comparison indicate that the spatial models perform better than the conventional generalized linear models. However, mixed results are obtained when the FBSA models were compared to the geographically weighted Poisson regression models.

Keywords:

• macro-level collision prediction models
• safety-planning models
• generalized linear models
• full Bayesian semiparametric additive models
• geographically weighted Poisson regression models
• Goodness-of-Fit measures