Modelling Motorway Accidents using Negative Binomial Regression

Abstract

This paper investigates motorway safety by developing accident prediction models that link accident frequencies to their non-behavioral contributing factors, including traffic conditions, geometric and operational characteristics of road, and weather conditions. The study used a sample of accidents occurred from 2004 through 2010 on a 74 km long section of Auckland motorway. A number of accident prediction models were developed and assessed for their predictive ability using negative binomial regression models under three categories: first for the whole of the motorway, second for rural and urban motorway segments separately and third for motorway segments without ramp, with on-ramp and with off-ramp separately. The results uncovered the safety impacts of different non- contributing factors, in which segment length, AADT per lane and the number of lanes always have the most profound effects on accident frequency. The validation tools were applied to examine the ability of models to predict accidents.