Predictive modeling to evaluate carrier safety performance in Saskatchewan

Author(s): Rajib Sahaji, George Eguakun, Cari Donaldson

Slidedeck Presentation:

4C Sahaji_Predictive Modeling to Evaluate Carrier Safety Performance in Saskatchewan

Abstract:

Carrier Safety performance in Saskatchewan is assessed on each of three safety measures: atfault
collisions, convictions and on road inspections, with each measure having its own unique
threshold. When a carrier’s performance on any of these three measures exceeds a threshold,
remedial action is triggered. However, this approach to assessing carrier safety performance
does not address the combined roles played by the three safety measures in predicting a
carrier’s future collision involvement. Using binary regression techniques, this study developed
predictive models to identify which past violation types significantly predict a carrier’s
involvement in future at-fault collisions, and, identify the specific types of collisions, convictions
and out-of-service (OOS) inspection violations that warrant attention when reviewing a carrier’s
safety performance. At the aggregate level, the model outcome indicated that a carrier with a
past at-fault collision, conviction, and OOS inspection violation is 3.62, 2.62, and 2.94 times as
likely as one with no violation history to be involved in a future casualty collision respectively. At
the disaggregate level, involvement in prior at-fault casualty collisions more significantly
determined a carrier’s likelihood of being involved in a future casualty collision than involvement
in past at-fault property damage collisions. Five out of 15 conviction types and two out of eight
OOS inspection types were found to be significant determinants of future at-fault casualty
collisions. This study provides the administrator with an empirical approach in determining the
relative importance of violation types in the carrier safety rating process by considering a
carrier’s propensity to be involved in future at-fault casualty collisions.