Predicting Driver Injury Severity in Single-Vehicle and Two-Vehicle Crashes with Boosted Regression Trees

The boosted regression tree model is an emerging nonparametric tree based model that can capture nonlinear effects of both discrete and continuous variables without preprocessing data. The model is particularly advantageous to predict severe injuries, which are more difficult to classify because of their small amount compared with nonsevere injuries. The objectives of this study were to investigate driver injury severity with the boosted regression tree model and other nonparametric models-the classification and regression tree and Random Forests-and to evaluate performance of the boosted regression tree model in comparison with the classification and regression tree model. The study identified important factors affecting injury severity by using 5-year crash records for provincial highways in Ontario, Canada. The results of the boosted regression tree model showed that ejection from a vehicle and head-on collisions commonly had a strong association with driver injury severity. Results also showed that marginal effects of continuous variables including truck percentage, annual average daily traffic (AADT), driver age, and vehicle age on injury severity were nonlinear. In particular, their effects on the injuries of heavy-truck drivers had different patterns compared with the effects on passenger-car and light-truck drivers; the risk of severe injury to heavy truck drivers increased as the truck percentage and AADT increased and the driver's age decreased. The boosted regression tree model predicted driver injury severity more accurately than the classification and regression tree model for both single-vehicle and two-vehicle crashes. Thus, it is recommended that the boosted regression tree model be applied with separate data sets for single-vehicle crashes and different types of two-vehicle crashes for more accurate prediction of crash injury severity.