This paper describes the development and assessment of a model for predicting accident frequency on rural single carriageway roads. A data base describing the accident frequency, layout, geometric design and traffic characteristics of 500 km sub-sections of single carriageway road has been assembled. The data base provided a wide range of values for each of the variables which were included. The microscopic simulation program OSCA was used to estimate measures of traffic performance for each of the 516 subsections in the data base. Alternative models were evaluated using Microsoft Excel and GLIM. A multiple linear regression model for accident frequency was found to provide the best fit to the data with the independent variables explaining 42% of the variation in accident frequency. Deviation from the mean speed for the road was the variable which provided the best improvement in fit and made the highest contribution to predicted accident frequency. The predicted accident frequency was also sensitive to the number of accesses present in a subsection. Models developed using a neural network were less effective in predicting accident frequency than the models developed using multiple linear regression. The model could be applied to evaluate variations in accident frequency which would be associated with changes in layout, geometric design, or traffic flow on rural single carriageway roads. (A)
Abstract