Research Papers

Transferability Assessment of Collision Prediction Models for Rural Highways in British Columbia

Version 1
Date added June 18, 2014
Downloaded 0 times/fois
Category 2014 CMRSC XXIV Vancouver
Tags Research and Evaluation, Session 4C
Author/Auteur Tarek Sayed, Emanuele Sacchi, Paul deLeur
Stream/Volet Research and Evaluation

Slideshow Presentation Only (no paper submitted)

4C Sacchi_Transferability Assessment of Collision Prediction Models for Rural Highways


The success of safety management programs in reducing traffic-related collisions depends on the existence of methods that give reliable estimates of road safety. In 2007, the British Columbia (BC) Ministry of Transportation and Infrastructure (MOTI) in partnership with the Insurance Corporation of BC (ICBC) completed a project to develop a set of collision prediction models (CPMs) for road segments and intersections in the Province. However, collision reporting practices have changed since then which requires updating existing BC models. The objective of this research was to investigate several options for BC CPMs calibration. These options included: the calibration and updating of existing models using BC data; the calibration and transferability of the Highway Safety Manual (HSM) models to be used in BC; or the development of new models for BC conditions.

The project was carried out using a set of recent collision and traffic volume data (2007-2011) for two-way two-lane rural highway segments. Two main calibration approaches were investigated. The first approach, proposed by the Highway Safety Manual (HSM), includes the use of a calibration factor. The second approach includes the transfer of the existing-model coefficients, by recalibrating only the constant of the model equation. The latter approach was developed at the University of British Columbia (UBC).

The study showed that developing new CPMs was a necessary task in order to update the existing BC models, which overpredicted the number of collisions by about 20% for the new time frame. However, existing BC models can still be recalibrated with the method described in the study. In fact, once they were recalibrated, the fit was adequate although not as good as estimating new models. This finding suggested that for some other model categories where data is difficult to collect (e.g., unsignalized intersections), existing BC models might be transferred.

It was also found that the HSM transferability method that makes use of the calibration factor was less reliable compared to the method developed at UBC. Moreover, the results showed that the fit of HSM models to BC data is generally poor. Therefore, the choice to use the HSM CPMs should be avoided for BC highways.

Tarek Sayed, Emanuele Sacchi, Paul deLeur