Investigating pedestrian crash injury patterns: A comparative study of children and non-children

Pedestrian injuries remain a public health concern, with child pedestrians being particularly vulnerable due to their unique physical and cognitive characteristics. This study presents a comprehensive analysis comparing injury severity patterns between child ($\le$14 years) and non-child ($>14$ years) pedestrians using Lasso logistic regression and advanced machine learning techniques, specifically Catboost with SHAP (SHapley Additive exPlanations) values to interpret the models. By analyzing six years of national crash data from the Crash Report Sampling System (CRSS) from 2016 to 2021, we identify significant factors influencing injury outcomes for both age groups. Our findings reveal that several variables were consistently associated with injury severity across both age groups and modeling approaches, including speed limit, lighting condition, pedestrian age, pedestrian contributing factor, vehicle pre-event movement, vehicle body type, traffic control, and intersection. However, key differences emerged. For child pedestrians, roadway surface and pedestrian position were identified only through a statistical modeling approach, while factors such as driver age, drug involvement, accompanying status, and year of crash were only found in the CatBoost model. In contrast, non-child models identified a broader set of driver-related factors, including age, drinking, and drug usage, which were less influential for children. The study demonstrates the value of integrating machine learning with traditional statistical methods to capture complex relationships and improve the understanding of injury severity for vulnerable road users. These findings offer valuable insights into pedestrian injury patterns across different age groups, informing targeted interventions aimed at enhancing pedestrian safety.