Citation

Dash I, Abkowitz MD, Philip C. J. Saf. Res. 2022; 83: 128-138.

Copyright

(Copyright © 2022, U.S. National Safety Council, Publisher Elsevier Publishing)

DOI

10.1016/j.jsr.2022.08.010

PMID

36481004

Abstract

INTRODUCTION: Bicycling plays an important role as a major non-motorized travel mode in many urban areas. While increasingly serving as a key part of an integrated transportation demand management system and a sustainable mobility option, interest in biking as an active transportation mode has been unfortunately accompanied by an increase in the number of bike crashes, many with incapacitating injuries or fatal outcomes. Thus, to improve bicycling safety it is crucial to understand the critical factors that influence severe bicyclist crash outcomes, and to identify and prioritize policies and actions to mitigate these risks.

METHOD: The study reported herein was conducted with this objective in mind. Our approach involves the use of classification models (logistic regression, decision tree and random forest), as well as techniques for treating unbalanced data by under sampling, oversampling, and weighted cost sensitivity (CS) learning, applied to bike crash data from the State of Tennessee's two largest urban areas, Nashville and Memphis.

RESULTS: The results indicate that random forest with weighted CS offers the potential for greater explanatory accuracy, an important observation given the paucity of efforts to date in applying random forest to bike safety studies. Inadequate lighting conditions, crashes on roadways, speed limits, average annual daily traffic, number of lanes, and weekends are the critical features identified.

CONCLUSION: Based on these results, a series of specific, suggested policy changes are presented for implementation consideration.

PRACTICAL APPLICATIONS: There is existing guidance in FHWA Lighting Handbook and TDOT's Roadway Design Guidelines that spell out some engineering design solutions like lighting provisions, bicycle facility design, and traffic calming measures. These measures may alleviate the identified key features impacting fatal and incapacitating bicycle injuries. Further research should be conducted to gauge the efficacy of the solutions suggested.

Language: en

Keywords

Policy; Safety; Bicycling; Machine learning; Sustainable transportation