Unraveling Pedestrian Fatality Patterns: A Comparative Study with Explainable AI

Age and substance use are critical risk factors: The study found that older adults and individuals under the influence of alcohol or drugs face significantly higher pedestrian fatality rates, suggesting that targeted interventions—such as enhanced safety education for seniors and substance abuse prevention programs—could save lives.

Location and visibility matter more than expected: Pedestrian deaths are more common in mid-block locations (between intersections) and areas with poor visibility, indicating that urban planning decisions about street lighting, sight lines, and pedestrian crossing infrastructure directly impact safety outcomes.

AI can reliably identify safety patterns: The XGBoost model achieved 90% accuracy in predicting contributing factors to pedestrian fatalities, demonstrating that machine learning can be a trustworthy tool for policymakers to identify which factors warrant intervention, though the study's focus on five high-fatality and five low-fatality states may limit broader applicability.

Explainable AI reveals actionable insights: By using SHAP values to interpret model decisions, researchers could identify why certain factors predict fatalities rather than just that they do, enabling policymakers to design evidence-based safety measures rather than relying on intuition alone.

Data-driven approaches can reduce disparities: The study's comparative analysis between high and low-fatality states suggests that examining what works in safer regions could help high-fatality states implement proven strategies, potentially addressing geographic inequities in pedestrian safety.