首页 >> 学术论文 >> 交通安全管理

Crash-type propensity analysis with bayesian models using microscopic traffic and weather data

发表时间:   阅读次数:

Rongjie Yu,Mohamed Abdel-Aty,Mohamed Ahmed,Xuesong Wang

Abstract:This study investigates a range of effects of microscopic traffic and weather factors and roadway geometry information on the specific crash type for a mountainous freeway. Crashes have been categorized as rear-end, sideswipe and single-vehicle crashes. Six-minute Automatic Vehicle Identification (AVI) segment average speed, real-time weather data and roadway geometry data are utilized as explanatory variables in this study. First, two binary logistic regression models were estimated by comparing single-vehicle to multi-vehicle crashes and sideswipe crashes to rear-end crashes. Then a full model which simultaneously fits two conditional logistic regression models (mixed logit model) for the three crash types has also been estimated. Results from the models indicate that single-vehicle crashes are more probable in the snow season, at moderate slopes, three-lane segments, under the free-flow conditions; while the sideswipe crash occurrence differs from rear-end crashes with the visibility situation, number of lanes, grades and their directions (up or down). Moreover, the results of the Bayesian random effects logistic regression models have been compared with the results from the classic logistic regression with the Frequentist and Bayesian inference techniques. It was demonstrated that the Bayesian random effects logistic regression outperforms the other two approaches with higher accuracy and lower Brier scores. The innovative way of estimating two conditional logistic regression models simultaneously in the Bayesian framework fits the data structure well. Conclusions from this study imply that different active traffic management strategies should be designed for three- and two-lane roadway sections and also considering the seasonal effects

©CopyRight 2003-2012   同济大学交通运输工程学院

备案号:沪ICP备1300535