Exploring Pedestrian Safety on Interstates with Bayesian and Machine Learning Models

Despite being prohibited from walking on freeways per federal laws, 14 to 17% of all pedestrian crashes in the United States happen on the interstates. Examining these crashes within the context of the safe systems approach is essential with an emphasis on the mitigation of safety risks for all road users. This study investigates the correlates of pedestrian crash injury severity on interstates in North Carolina, focusing on pedestrian actions, roadway conditions, and the type of vehicles involved in the crashes. The study utilizes police-reported pedestrian crash data from 2007 to 2022, coded by the Pedestrian and Bicycle Crash Analysis Tool (PBCAT) which provides unique and comprehensive crash descriptors. The analysis considers 882 pedestrian crash observations on freeways. The dependent variable, pedestrian injury severity, is categorized into distinct binary outcomes, fatal and severe injuries versus minor injuries. The study applies frequentist and Bayesian binary logit models with various prior specifications, along with a robust machine learning algorithm Random Forest for their ability to provide reliable estimates even with a limited sample size. The results show that pedestrian crashes are more predominant in rural (47%) than urban freeways (40%) in North Carolina. Key findings indicate that pedestrians crossing the roadway, crashes on dark unlit roads, and crashes involving pedestrian alcohol impairment are associated with a higher risk of severe injuries to pedestrians. The study findings highlight the need for faster roadside assistance to stranded pedestrians, stringent penalties for alcohol impairment, and the implementation of physical barriers preventing pedestrian access on freeways.