Identifying Factors That Increase Bus Accident Risk by Using Random Forests and Trip-Level Data

Abstract

Responsible bus accidents—accidents in which an operator’s actions largely explain why an accident occurred—decrease service reliability and safety, cause property damage, and lead to injury or even death. These effects make the reduction of avoidable accidents a high priority for transit authorities. This study used innovative techniques to identify factors that affected the probability of a responsible bus accident in the Minneapolis–Saint Paul, Minnesota, metropolitan area. The study examined data on the trip level and used the random forest model to identify the factors that most strongly affected the likelihood of an accident. In addition, this study used the partial dependency function of the random forest model as an aid in identifying nonlinear relationships between predictor and response variables. This function informs variable transformation in a traditional logistic model, which accurately classifies test set trips as accidents and nonaccidents at a rate of about 72%. Many results fell in line with past literature, but new insights were gained. Specifically, in keeping with most past studies, accidents were more likely when the bus operator was older, female, fatigued, inexperienced, driving a larger bus, or driving on a dense urban route. One new insight of this study is that bus drivers are at greater risk toward the middle of their shift, especially when traffic is dense. Bus drivers’ risk greatly increases if they ( a) did not work the previous day and ( b) worked longer hours the previous week. Because previous studies have not examined accidents on the trip level, they have not found these more specific work-hour relationships that capture elements of alertness, fatigue, and complacency.