HubkaRao, Tate; Nettel-Aguirre, Alberto; Cloutier, Marie-Soleil 
ORCID: https://orcid.org/0000-0002-8533-4784 et Hagel, Brent E.
(2026).
Using machine learning to predict child active transportation related motor-vehicle collisions
Journal of Transport & Health
, vol. 49
, nº 102315.
DOI: 10.1016/j.jth.2026.102315.
Résumé
Background: Motor-vehicle collisions (MVCs) are a leading cause of child injuries in Canada.
Children have unique active transportation patterns; however, little research exists predicting the
occurrence of MVCs with child active transportation users. Further, current machine learning
prediction models can lack interpretability, hindering usability for built environment change.
Objective: Develop a novel and interpretable machine learning recursive partitioning tree to
predict MVC injury related to child active transportation.
Methods: The Child Active-Transportation Safety and the Environment study's geodatabase,
including population demographics, built and school environment, mode of transportation, and
MVCs, was used to train a Poisson Regression tree. Data were collected from five Canadian
municipalities/regions and aggregated by dissemination area (DA). Both national and cityspecific models were trained, with the data for each model separated into training (80%) and
validation (20%) datasets. Root mean squared error (RMSE) was used to assess prediction accuracy and variable importance was also calculated.
Results: Over 10,000 DAs were included. National level models had the best accuracy with a
relative RMSE of 6. City-specific models ranged in accuracy from 7 (Calgary and Peel Region) to
10 (Laval). The number of signalized intersections and length of major roads were frequently
ranked high in importance for prediction of child bicyclist and pedestrian-MVCs, across national
and city models.
Conclusions: Prioritizing safety at signalized intersections and major roadways may prove most
effective for interventions at specific locations. The complexity in both national and city-specific
trees suggests a multi-pronged strategy is best overall, in alignment with current Safe Systems
approaches to road safety
| Type de document: | Article |
|---|---|
| Mots-clés libres: | machine learning; child safety; built environment; motor-vehicle collision; injury prevention |
| Centre: | Centre Urbanisation Culture Société |
| Date de dépôt: | 30 avr. 2026 19:26 |
| Dernière modification: | 30 avr. 2026 19:26 |
| URI: | https://espace.inrs.ca/id/eprint/17092 |
Gestion Actions (Identification requise)
 |
Modifier la notice |