NETWORK-LEVEL VEHICLE CRASH PREDICTIONS INCORPORATING TIME-DEPENDENT EFFECTS INTO CONSIDERATIONS
MetadataShow full item record
Maintaining highway safety is viewed as the over-arching goal of mananging transportation systems at all levels. According to the National Highway Traffic Safety Administration (NHTSA), over 37,000 people got killed and 2.35 million are injured in road crashes annually. The equivalent economic and societal losses are on the order of over $231 billion, or an average of $820 per person. Thus, developing vehicle crash models that can accurately predict crash occurrences becomes essential. The study begins with literature review of models for predicting vehicle crash frequencies and crash severity levels on highway segments and at highway intersections. The findings of literature review indicate that some models lack prediction accuracy owing to exclusion of many crashing contributing factors. Consequently, a new methodology for improved vehicle crash predictability is proposed to include as many crash contributing factors as possible. In addition, the proposed methodology aims to conduct crash predictions targeting a highway network. Two computational experiments are performed for methodology application, including one on highway segment-related crash predictions using data on Highway Safety Information System for Illinois from 2001 to 2010, and another one on intersection-related crash predictions using crash data on more than one thousand intersections for period 2004 to 2010 provided by city of Chicago. Cross comparisons are made on the results obtained by applying the proposed methodology, method documented in Highway Safety Manual (AASHTO, 2010), and Empirical Bayesian (EB) before-after method for validation. The proposed methodology is found to have out-performed the other two methods. Future research directions are provided for continuing refinements of the proposed methodology.