Abstract for presentation (Poster or Podium)
Transportation Safety
Manmohan Joshi, n/a
Graduate Student
University of Connecticut
Mansfield Center, Connecticut, United States
John Ivan, n/a
Professor of Civil Engineering
University of Maine
Storrs, Connecticut, United States
Kai Wang, n/a
Assistant Research Professor
University of Connecticut
Storrs, Connecticut, United States
Shanshan Zhao, n/a
Research Scientist
University of Connecticut
Storrs, Connecticut, United States
Tanmoy Bhowmik (he/him/his)
Assistant Professor
Portland State University
Portland, OR, United States
Shahrior Pervaz, n/a
Graduate Student
University of Central Florida
Orlando, Florida, United States
Naveen Eluru, n/a
University of Central Florida
Orlando, FL, United States
Manmohan Joshi
University of Connecticut
Mansfield Center, Connecticut, United States
Accurate crash prediction models are essential for road safety and management. This study focuses on the impact of vehicle mix, specifically the presence of trucks, on crash prediction. Trucks can induce discomfort, reduce visibility for other drivers, and create larger blind spots, increasing the risk of accidents. However, obtaining and maintaining comprehensive data with classified volume counts for all network facilities is resource intensive. To overcome data challenges, the Quasi-Induced Exposure (QIE) method is employed to estimate truck proportions for modeling crash severity. The research compares the accuracy of crash severity prediction models using vehicle mix data from observed or estimated traffic counts, QIE-generated data, and models without vehicle mix data. The research utilizes data from the Highway Safety Information System (HSIS), which was collected over a five-year period (2013-2017) for urban freeway segments in California. A Multivariate Poisson Lognormal (MVPLN) model with random effects is used, estimated through Markov-Chain Monte-Carlo Simulations. This model considers correlations among crash severities at the site level and accounts for heterogeneity due to heavy vehicles. The results highlight the significant influence of truck proportions on crash severity predictions. Furthermore, the study suggests that QIE-generated vehicle mix data is a reliable substitute for observed data in crash prediction models. This research underscores the importance of considering vehicle mix, especially trucks, in crash prediction models. It introduces QIE as a valuable method for estimating this data and suggests its effectiveness in improving the accuracy of crash prediction models.