Abstract
This study investigates the frequency and injury severity of pedestrian crashes across Texas using tree-based machine learning models. Ten years of police records are used along with roadway inventory and other sources to map more than 78,000 pedestrian crashes over 700,000 road segments along with road design, land use, transit stops,and hospital location and weather information. Methods such as random forests (RF), gradient boosting (LightGBM and XGBoost), and Bayesian additive regression trees (XBART) are applied and compared. The crash frequency models indicate that highway design variable significantly positively impact pedestrian crash frequencies. Increments in total or fatal crash counts are related to the higher number of lanes, while higher speed and median and shoulder widths lead to fewer crash frequencies. Other variables such as proximity to schools, the number of transit stops, and population and job density increased pedestrian crash occurrences. Pedestrian severity models found that speed limit significantly increases the likelihood of pedestrian fatalities and severe injuries, and intoxicated drivers and pedestrians lead to more severe injuries. Also, pedestrian crashes are more likely to be severe and fatal at night and in areas with poor lighting conditions. An analysis of the vehicle type found that light-duty trucks (pickups, SUVs, and vans) also increase pedestrian severity. The comparison of the four models indicates that they performed similarly in predicting crash occurrences, with LightGBM showing significantly lower computational time. While for crash injury severity models, XBART obtained a higher precision value but with a significantly high computational time.
References: Bo Zhao, Natalia Zuniga-Garcia, Lu Xing and Kara Kockelman. Predict Pedestrian Crash Occurrence and Injury Severity in Texas Using Tree-Based Machine Learning Models. https://www.caee.utexas.edu/prof/Kockelman/public_html/TRB22MLPedCrashes.pdf