Exploring Pedestrian Injury Severity by Incorporating Spatial Information in Machine Learning

Abstract

Using the random forest classification technique, this study explored the role of different factors such as demography, pedestrian and drivers' conditions, collision characteristics, road characteristics, and weather in predicting pedestrian injury severity from automobile-related collisions in Toronto. Spatial information was incorporated in the models to capture spatial autocorrelation. The results revealed the importance of spatial information in predicting pedestrian injury severity. Other important predictors of pedestrian injury severity include aggressive driving, driver's conditions (e.g., inattentive, slowly stopping, driving properly, failing to yield right of way), pedestrian conditions (e.g., normal, inattentive) and dark lighting conditions.