Predictive models for road accidents at signalised intersection

Large numbers of accidents occur at intersections in urban street networks. The major intersections most often controlled by signals contribute to the majority of these collisions. While the reason for these are varied and complex, a significant part may be attributed to the built environment in terms of both the physical and traffic management systems. This paper reports on the development of appropriate predictive models for accidents at signalised intersections capable of being used in planning purposes. Four years of accident data at 115 signalised intersections are assembled in a database using a Geographic Information System (GIS). Non-linear and stepwise multiple linear regression methods are used to develop models for each of the main types of accidents at signalised intersections in South Australia. It is known that different traffic flow movements lead to different types of accidents. Hence for each accident type, the accident frequency was related to a functional form of the traffic flow movements contributing to that type of collision, signal parameters and intersection geometry and location factors. The study found that: 1) Valid predictive models can be developed by relating the various types of accidents to the traffic movements contributing to their occurrences. 2) The various types of accidents relate to different functional form of the flow exposure measure, site and signal parameters, so that models based on aggregate data are inappropriate. 3) Differences exist between the characteristics and number of accidents occurring at signalised intersections in the Adelaide CBD and those outside the CBD due to differences in land use activities and signals operation.