Bayesian analysis of the effect of horizontal curvature on truck crashes using training and validation data sets

Abstract: The research on the effects of roadway geometry on truck crashes is relatively limited in comparison with predictive models developed for total vehicle crashes. The most common predictive models currently used are Poisson and negative binomial models. This study uses a negative binomial model but applies the full Bayes’ methods for improving model performance. To use Bayes’ methods successfully, a learning process was used to develop a final model, which was then compared with a separate validation data set to verify its accuracy. The data set used for this study is based on rural two-lane collector and arterial horizontal curves in Ohio, comprising 15,390 observations from crash records between 2002 through 2006. Specific areas of interest in this study include the impact of shoulder width, horizontal curve radius, curve length, and other traffic parameters. The final results indicate a significant increase in truck crashes due to both horizontal curvature and passenger vehicle volumes. The final model’s predictions were improved compared with the initial model, indicating that the learning process is a viable tool for future crash model development.