Comprehensive Investigation of Pedestrian Hit-and-Run Crashes: Applying XGBoost and Binary Logistic Regression Model

Ahmed Hossain,Xiaoduan Sun,Ahmed Sajid Hasan,Mohammad Jalayer,Julius Codjoe
DOI: https://doi.org/10.1177/03611981241262315
2024-01-01
Transportation Research Record Journal of the Transportation Research Board
Abstract:The present trend in the United States suggests that one in five pedestrian fatalities in motor vehicle crashes involves a hit-and-run, a serious traffic safety concern. The over-representation of pedestrian hit-and-run collisions necessitates a systemic data-driven investigation to uncover the contributing factors that cause fatalities or serious injuries. This study addressed two research questions (RQ), RQ1: What factors contribute to pedestrian hit-and-runs? RQ2: What causes hit-and-run pedestrian fatalities? This study addresses the RQs using the XGBoost algorithm (RQ1) and binary logistic regression model (RQ2) to analyze police-reported pedestrian crashes (2015–2019) in Louisiana. The XGBoost model was used to classify pedestrian hit-and-run crashes (hit-and-run = yes/no) and identified critical factors as predictors of pedestrian hit-and-run crashes including: primary contributing factors (pedestrian action, pedestrian violation, prior movement, pedestrian condition); settings (dark-with-streetlight, posted speed limit of > 55 mph, two-way road with physical separation); pedestrian characteristics (younger and older pedestrians, male gender, presence of dark clothing); and weekend. The binary logistic regression model was further used to identify critical high-risk hit-and-run scenarios resulting in fatal or severe injury of pedestrians. Some of the identified top factors are posted speed limit of 55 mph or higher (OR = 12.74), pedestrian impairment (OR = 4.77), older pedestrians (OR = 2.68), younger pedestrians (OR = 1.79), and dark-no-streetlight conditions (OR = 2.91). Both models showed strong relationships between pedestrian hit-and-run crashes and fatal or severe injuries (e.g., dark-with-streetlight, high-speed settings, older pedestrians, and pedestrian actions). Identifying these critical links can help policymakers, law enforcement agencies, and transportation authorities develop targeted interventions and strategies to address the risk factors.
What problem does this paper attempt to address?