Nighttime pedestrian safety in different communities: Application of artificial intelligence techniques

Area of Science:

• Transportation Safety
• Public Health
• Data Science

Background:

• Pedestrian safety is a critical concern, with thousands of fatalities annually in the U.S.
• Nighttime crashes pose higher risks due to reduced visibility and impairment.
• The U.S. Department of Transportation uses six indicators to enhance safety.

Purpose of the Study:

• To integrate U.S. Department of Transportation indicators with pedestrian crash data.
• To identify key correlates of nighttime pedestrian crash injury severity.
• To develop an accurate predictive model for pedestrian crash outcomes.

Main Methods:

• Utilized an inference-based ordered logit model for analysis.
• Applied an Artificial Intelligence (AI) based heterogeneous ensemble method (Stacking).
• Employed machine learning techniques including Gradient Boosting, Decision Tree, and Random Forest.

Main Results:

• The stacked AI model achieved 78.85% predictive accuracy, outperforming base learners.
• Identified associations between socioeconomic indicators (Economy, Transportation Access) and crash severity.
• Found links between unlit roads, crossing violations, alcohol impairment, and severe nighttime pedestrian crashes.

Conclusions:

• AI-driven insights can significantly aid safety practitioners.
• Findings support targeted interventions to improve roadway infrastructure.
• The study contributes to enhancing overall road user safety through data-driven strategies.