Nighttime Pedestrian Safety in Burdened Communities: Application of Artificial Intelligence Techniques

Pedestrian safety is a growing concern in the US transportation sector, with around 7500 pedestrian crash fatalities reported in 2021. Already highly susceptible to traffic crashes, pedestrians are at an even higher risk of crashes at night. This study integrates six unique transportation burden indicators—Economy, Health, Equity, Resilience, Environmental, and Transportation Access—developed by the United States Department of Transportation at the census tract level with nighttime pedestrian crash data from 2016-2019 in North Carolina. The pedestrian crash data are extracted from police reports using the Pedestrian and Bicyclist Crash Analysis Tool, which provides high-quality detailed crash-type descriptors, resulting in a unique and comprehensive pedestrian crash database. The study applies rigorous methods for analysis, including the inference-based ordered logit model, to quantify key correlates of nighttime pedestrian crashes in burdened communities (BCs). The model results reveal unique and novel associations of the Economy and Transportation Burden indicators, roads without lights, pedestrian crossing violations, and alcohol impairment with nighttime pedestrian crash injury severity. To improve forecasting of pedestrian crashes and the resulting injury severity in BCs for planning purposes, an Artificial Intelligence (AI) based heterogeneous ensemble method, “Stacking,” is applied with an Ordered Logit model and machine-learning techniques, Gradient Boosting, Decision Tree, and Random Forest as the base learners. The stacked model yields better predictive accuracy than the individual base learners. The study findings and the application of AI techniques can assist safety practitioners in improving planning and implementing targeted interventions in BCs to improve roadway infrastructure and overall safety.