A model to predict the probability of highway rail crossing accidents

Abstract

In the current paper, a model to predict the probability of accidents, injuries, and fatalities resulting from collisions between trains and vehicles at highway rail crossings is presented. Logistic regression and two databases maintained by the US Federal Railroad Administration (FRA) are used to build the models. These models prove more than an order of magnitude more accurate than a previous model developed by the FRA, which is currently used to compute the cost effectiveness of crossing upgrades. A declining trend in the likelihood of an accident that cannot be explained by changes in crossings or highways over time is discovered. Possible causes of this trend and test one of these possibilities are discussed. The model is used to compute both the cost per life saved from upgrading each crossing without gates and the trend over time in the number of crossings that are cost effective to upgrade.