A machine learning-supported path analysis to uncover the behavioral pathways in pedestrian-involved traffic crashes

Pedestrians are vulnerable road users, and many studies have examined the characteristics of crashes involving pedestrians, aiming to address the factors contributing to pedestrian injuries and fatalities in traffic crashes. Unlike existing studies that directly link factors to pedestrian injuries, this study aims to uncover the behavioral pathways in traffic crashes involving pedestrians. It is based on the assumption that a pedestrian's pre-crash behavior is the outcome of multiple contributing factors, including pedestrian demographics, traffic conditions, and environmental characteristics; and also, a pedestrian's risky behaviors, such as failing to yield or dash/dart-out, can lead to severe injuries in traffic crashes. Therefore, a behavioral pathway can be formed between the contributing factors, pedestrian pre-crash behaviors, and pedestrian injury outcomes in traffic crashes.

Using data on pedestrian-involved crashes from 2018 to 2022 in North Carolina, this study employs a path analysis framework integrated with interpretable machine learning models to examine the behavioral pathways in pedestrian-involved crashes. The path analysis allows the identification of various factors that directly contribute to injury severity and those that indirectly contribute to pedestrian injuries through their pre-crash behaviors.

The results indicate that several factors are directly associated with pedestrian injury severities, including pedestrian demographics, pre-crash behaviors, vehicle features, driver's intoxication, and road environment. Further, some factors serve as predictors of pedestrians' pre-crash behaviors, indirectly contributing to their injury severity.

This study provides insights into the behavioral pathways leading to pedestrian injuries, informing educational campaigns, infrastructure improvements, and enforcement strategies to reduce pedestrian injuries and fatalities.