This paper describes a novel method for the analysis of road accident data extracted from STATS/19 records. The objective of this approach is to perform an automatic first scan of the accident record of a site and to identify particular features which appear, on the basis of this record, to be unusual, and to quantify the strength of this identification. This information is then used to suggest candidate accident remedial measures for further consideration. The main feature of this approach is the integration of empirical Bayesian statistical techniques and Bayesian inference, with uncertain inference within a knowledge-based system. The integration of these techniques is in accordance with the character of road accident remedial work which requires inferences to be made on the basis of uncertain evidence derived from historical records of a relatively small number of road accidents which have occurred at an individual site to be treated. The system has been applied to a real road accident database including more than ten thousands of accidents. The preliminary results have shown that the approach is relevant and useful. One of the examples is a site which was investigated by engineers of the London Accident Analysis Unit (LAAU) who identified conversion from priority control to a mini-roundabout as being suitable treatment. On the basis of the accident history, the system also identified this conversion as being one of a number of appropriate measures.
Samenvatting