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

Pedestrian pre-crash behaviors such as failing to yield and dash/dart-out are significant mediators between contributing factors and pedestrian injury severity.

• The study assumes that pre-crash behaviors are outcomes of multiple contributing factors including pedestrian demographics, traffic conditions, and environmental characteristics
• Risky behaviors including failing to yield or dash/dart-out were found to lead to severe injuries in traffic crashes
• The path analysis framework explicitly models pre-crash behaviors as intermediate variables between upstream factors and injury outcomes
• Data from pedestrian-involved crashes from 2018 to 2022 in North Carolina were used

Several factors were found to be directly associated with pedestrian injury severities in traffic crashes.

• Factors directly contributing to injury severity include pedestrian demographics, pre-crash behaviors, vehicle features, driver's intoxication, and road environment
• These direct associations were identified through the path analysis framework independent of behavioral mediation pathways
• The study used an interpretable machine learning model integrated with path analysis to quantify these direct associations

Some factors serve as predictors of pedestrians' pre-crash behaviors, thereby indirectly contributing to pedestrian injury severity.

• The path analysis framework allowed identification of factors that indirectly contribute to pedestrian injuries through their influence on pre-crash behaviors
• This indirect pathway mechanism distinguishes this study from existing research that directly links factors to pedestrian injuries
• Contributing factors identified as predictors of pre-crash behavior included pedestrian demographics, traffic conditions, and environmental characteristics

The study employed a path analysis framework integrated with interpretable machine learning models to examine behavioral pathways in pedestrian-involved crashes.

• The dataset covered pedestrian-involved crashes from 2018 to 2022 in North Carolina
• The integrated framework combined path analysis with interpretable machine learning models
• The methodology allowed simultaneous examination of direct and indirect (behavior-mediated) pathways to injury severity
• The approach differs from conventional studies by explicitly modeling pre-crash behavior as an intermediate outcome rather than a direct predictor

The findings provide a basis for targeted interventions including educational campaigns, infrastructure improvements, and enforcement strategies to reduce pedestrian injuries and fatalities.

• Insights into behavioral pathways inform distinct types of interventions targeting different points in the causal chain
• Educational campaigns can target risky pedestrian pre-crash behaviors such as failing to yield and dash/dart-out
• Infrastructure improvements and enforcement strategies are implicated based on the road environment and traffic condition findings