Railroad Grade Crossings: Accident Prediction Models and Adjustments for Historical Data

Thursday, April 26, 2018

About the Presentation

This presentation explained how researchers used a zero-inflated negative binomial (ZINB) model to predict the accidents at rail grade crossings. The goal of the project was to improve predictions over the current state-of-practice USDOT accident prediction formula. As part of this work, three ZINB models were generated for three different categories of warning devices: gates, flashing lights, and crossbucks. Like the USDOT formula, the direct predictions from the models were adjusted using accident history data. Empirical Bayes (EB) adjustment was applied to ZINB models to improve the predictions.

Data for Illinois were used to develop the models, and data for four other states were used for model validation. Different comparisons were then made between the models. The comparisons showed that the ZINB model with EB adjustment outperformed other models because it had a cumulative accident distribution that closely represents the field data. Plots of actual accident counts versus predicted accident counts for each of the models showed that the EB-adjusted accident prediction value is closer to the actual accident counts than the other models. Also, more accurate predictions from the EB-adjusted ZINB model were observed for the top 10, 20, 30, 40, and 50 locations with highest accident frequency from each warning device category. In each of the three comparisons, the EB-adjusted ZINB model outperformed the USDOT model.

Webcast

About the Speaker

Ray Benekohal is a professor in the Department of Civil and Environmental Engineering at the University of Illinois Urbana-Champaign. He has conducted numerous studies on the development, evaluation, and analysis of transportation systems. His research focuses on traffic flow modeling and simulation, traffic flow theory, intelligent transportation systems, traffic operations, and traffic safety.