Analysis of the Safety Characteristics of Unsignalized Intersections

This study is concerned with using multiple approaches for analyzing safety at unsignalized intersections. This was investigated by analyzing total crashes, one of the most frequent crash types at unsignalized intersections (angle crashes) and crash injury severity. Some of the developed methodological techniques in this study are considered recent and have not been extensively applied. Massive data collection effort was conducted, and resulted in collecting around 2500 unsignalized intersections from six counties in Florida. The first analysis dealt with applying the Bayesian updating concept for better crash prediction. This was investigated by updating the coefficients of the probabilistic negative binomial (NB) models. Different non-informative and informative prior structures using the NB and log-gamma distributions were attempted. The log-gamma distribution showed the best prediction capability. The second analysis dealt with analyzing crash injury severity using the binary probit framework. The important factors found in the fitted probit models were the logarithm of the Analysis of the Safety Characteristics of Unsignalized Intersections Abdel-Aty, Mohamed; Haleem, Kirolos 12 th WCTR, July 11-15, 2010 – Lisbon, Portugal 2 annual average daily traffic “AADT” on the major road, and the speed limit on the major road. It was found that higher severity probability is always associated with a reduction in AADT, as well as an increase in speed limits. Finally, a recently-developed data mining technique (the multivariate adaptive regression splines or MARS) was used, which is capable of yielding high prediction accuracy. MARS was used to analyze angle crashes. MARS yielded the best prediction performance while dealing with continuous responses. Additionally, screening the covariates using random forest technique before fitting a MARS model was very encouraging.