Injury severity models for motor vehicle accidents: a review

Modelling of traffic accidents injury severity is a complex task. In the last few years the number and variety of studies that analyse injury severity of traffic accidents have increased considerably. In this paper 19 modelling techniques used to model injury severity of traffic accidents where at least a four-wheeled vehicle is involved have been analysed. The analysis and the comparison between models was performed based on seven criteria (modelling technique, number of records, number of variables, area type, features, injury level and model fit). In general, it is not possible to recommend a method that could be identified as the best one. Each modelling technique has its own limitations and characteristics, awareness of which will help analysts to decide the best method to be used in each particular modelling problem. However, some general conclusions can be established: in most cases the results of models' fits are found to be satisfactory, though not excellent; in the case of data mining models, accuracy improves with balanced datasets; and no correlation was found to exist between the number of accident records and the number of analysed variables.