Applying Artificial Intelligence techniques to examine nighttime pedestrian crash injury severity at intersections
Sheikh Muhammad Usman, Asad J. Khattak
Abstract
Efficient intersection design requires balancing road user safety and mobility. Due to minimal protection, pedestrians are especially vulnerable to crashes at intersections. In the U.S., pedestrian fatalities at intersections have been rising, with approximately 75% occurring at nighttime. This study identifies factors associated with nighttime pedestrian crash injury severity at intersections. It analyzes police-reported pedestrian crashes in North Carolina from 2016–2022, comprehensively coded using the Pedestrian and Bicyclist Crash Analysis Tool, which offers detailed descriptors and crash types for robust analysis. The study employs both statistical and Artificial Intelligence (AI) methods. Ordered Logit Model is estimated to quantify the impact of the correlates of pedestrian crash injury severity, while AI algorithms—XG Boost, Light GBM, and CatBoost—are used to predict pedestrian crash injury severity and support effective urban planning. Findings reveal significant associations between pedestrian injury severity and several behavioral, infrastructural, and regulatory factors at intersections, including pedestrian dash/dart-out behavior, drivers failing to yield, inadequate lighting, and high-speed limits. The findings can potentially lead to the development of AI-driven intersection safety systems aimed at detecting and mitigating unsafe pedestrian-vehicle interactions by leveraging real-time sensor data.