Modelling of Traffic Accidents on Motorway

Abstract

Traffic safety on motorway continues to be of great interest, as motorway carries busy vehicular traffic and experience a considerable number of accidents every year. Accident prediction model has been employed as a popular mathematical tool to estimate road safety for many years. The main objective of this thesis is to explore the knowledge related to motorway accidents by investigating the accident features and developing statistical models that link accident frequencies to their contributing non-behavioural factors, including traffic conditions, road geometric and operational characteristics, and weather conditions. A 74 km long section of Auckland motorway along the State Highway 1N was selected for this study considering only accident cases recorded from 2004 through 2010. A descriptive analysis on accident data revealed some general characteristics of motorway accidents in terms of time trend, crash type and movement, road and environment condition, and driver group. Among a variety of modelling approaches, the negative binomial distribution was selected for being capable of describing the random, discrete and non-negative accident events as well as accommodating the over dispersion that exists in accident count data. The thesis presents a detailed discussion of the problems related to the identification of significant explanatory variables out of potential candidates. Accident frequency models are developed for different accident datasets. On the issue of improving accident prediction accuracy, the thesis proposes to develop separate models for accidents on motorway segments with on-ramp, with off-ramp and without ramp. The results reveal that some explanatory variables play different roles on accident occurrence for models by motorway segment type or by accident category. In addition, segment length, AADT per lane and the number of lanes always have the most profound effects. The findings make the recommendation of effective countermeasures on motorway safety to be possible.