Accident Prediction Models for Hungarian Two-Lane Rural First-Class Main Roads

Abstract

This paper describes a modeling effort to define accident prediction models for first-class main roads outside built-up areas in Hungary using variables that are available and believed to exert an influence on safety performance. The first part of the paper describes data collection and the segmentation technique. Six years of accident data are gathered for segments only; intersections with a 200m radius are taken out in order to avoid having intersection related crashes in the data. Altogether 1357 homogenous sections are formulated based on AADT (Annual Average Daily Traffic), road width, posted speed, horizontal curve and shoulder width. Models are proposed using the Generalized Linear Modeling (GLM) approach assuming a negative binomial error structure. It is concluded that AADT, roadway width, horizontal curve and segment length significantly influence accident frequency. Estimated model parameters are explained by putting them into international context and it is concluded that the results are in accordance with previous research findings.