Applying Artificial Intelligence to Examine Nighttime Pedestrian Crash Injury Severity at Intersections

Intersections require a trade-off between road user safety and mobility. Due to minimum protection, pedestrians are highly vulnerable to traffic crashes at such locations. Recently, fatal pedestrian crashes at intersections have risen in the US, and about 75% of pedestrian fatalities occurred at nighttime. To enhance the safety of pedestrians at intersections, this study identifies the correlates of nighttime pedestrian crash injury severity at intersections. The study examines police-reported pedestrian crashes in North Carolina from 2016-2019, recoded comprehensively using the Pedestrian and Bicyclist Crash Analysis Tool. The tool provides a multitude of crash descriptors and crash types, resulting in a unique multi-faceted pedestrian crash database. The analysis involves estimating rigorous statistical models and innovative application of Artificial Intelligence tools. An Ordered Logit Model is estimated to quantify the correlates of nighttime pedestrian crash severity at intersections. The modeling results reveal interesting and novel insights into the association of several behavioral, infrastructural, and regulatory factors, including pedestrian dash or dart-out behavior, drivers not yielding to pedestrians, inadequate lighting, and high speed limit at intersections with pedestrian injury severity, given a crash. The study applied a powerful prediction-based AI algorithm, Random Forest, to obtain accurate forecasts of pedestrian crash injury severity for devising effective urban planning strategies and infrastructure improvements to mitigate pedestrian crash injury severity at intersections. The study can assist in realizing USDOT’s vision to develop an AI-based intersection safety system to anticipate, identify, and mitigate unsafe pedestrian-vehicle interactions at intersections by harnessing real-time information collected through emerging sensors.