A Misclassification Framework for Identifying Systematic Error in Pedestrian Crash Prediction: Evidence from Seattle,...

Area of Science:

• Transportation Engineering
• Urban Planning
• Data Science

Background:

• Pedestrian crash prediction models frequently exhibit systematic failures, making it hard to differentiate correctable errors from random noise.
• Accurate prediction is crucial for effective traffic safety interventions and urban planning.

Purpose of the Study:

• To develop and validate a framework for identifying systematic failures in pedestrian crash prediction models.
• To pinpoint high-priority locations requiring focused safety improvements by distinguishing correctable model errors.

Main Methods:

• A novel framework integrating spatial clustering, temporal persistence analysis, and model ensemble consensus was developed.
• The framework was applied to 13,706 Seattle intersections to identify locations with high systematic error.
• A multiplicative priority score was used as a proxy for systematic error, concentrating errors at specific locations.

Main Results:

• The framework identified 141 high-priority locations (1.03% of total) with significant temporal persistence, spatial clustering, and high ensemble consensus.
• These locations concentrated 6.6% of the total model error, indicating systematic prediction failures.
• Audits revealed omitted features like visual obstructions, static road use, and network complexity at these sites.

Conclusions:

• The developed framework effectively identifies high-priority locations with systematic errors in pedestrian crash prediction models.
• Adding omitted site-specific features significantly reduced error scores at identified locations, validating the framework's utility.
• This approach guides targeted interventions by pinpointing where missing contextual data most impacts crash prediction accuracy.