Citation

Nishimoto T, Kubota K, Ponte G. Accid. Anal. Prev. 2019; 129: 84-93.

Affiliation

Centre for Automotive Safety Research, The University of Adelaide, Adelaide, South Australia, Australia.

Copyright

(Copyright © 2019, Elsevier Publishing)

DOI

10.1016/j.aap.2019.04.021

PMID

31128444

Abstract

The purpose of this study was to develop a serious injury risk prediction algorithm for pedestrians, using data from the South Australian Traffic Accident Reporting System. Two algorithms were developed to estimate serious injury risk, using a logistic regression analysis of 6,868 vehicle-pedestrian crashes extracted from TARS data. In this study, an optimal model based on the best combination of risk factors according to the Akaike information criterion (AIC) was developed. Additionally, a secondary GPS model using only crash site characteristics that can be derived from GPS coordinates from the crash scene was also developed. The optimal model is based on site and environmental conditions that could be derived from GPS data (posted speed limit, distance from crash site, natural lighting conditions, road geometry, road horizontal alignment and road vertical alignment) as well as pedestrian age/gender, driver age/gender and vehicle model year. The second model only included features that could be derived from GPS data. The optimal model was reasonable in accuracy and gave an under-triage rate of 10% when the injury threshold was set to 15%, with a corresponding over-triage rate of around 60%. The GPS model, despite not being as accurate as the optimal model may be adequate in the absence of all the risk factors required for the optimal model, requiring an injury threshold of 20% to give an under-triage rate of 10%, with the corresponding over-triage rate being around 70%. Both models can potentially be used for serious injury risk prediction (SIRP) for pedestrians involved in a collision with a vehicle.

Copyright © 2019. Published by Elsevier Ltd.

Language: en

Keywords

Advanced automatic collision notification; Injury prediction; Logistic regression; Pedestrian injury; Prehospital care; Speed limit