Driver Pattern Identification in Road Crashes in Spain

Abstract

Extracting driver collision patterns by gender and age regarding offences, collision type and injury severity is very useful in road safety, providing a better understanding on behavior of the different driver groups. Self-Organizing Map (SOM) is the tool proposed for distributing and projecting 145,904 drivers according to 8 offence variables on a 2D map. Thus, drivers who are close in the original 8D space (one dimension per offence variable), will remain so in the projected one (2D). Multivariate driving and collision patterns are explored to support the development of future measures to improve road safety. Tests of proportions are used for shedding light on clusters where driver offence is present. Finally, the SOM results were compared for validation with those of the standard K-Means clustering technique. The results show that the characteristics of road crashes and the severity of injuries depend jointly, i.e., in multivariate (pattern) terms, on gender, age, type of collisions and offences. There are relevant multivariate driver behavior differences in both the type of collisions (and therefore their severity) and the type and number of offences with regard to gender and age of the driver. This research unveils different multivariate driver behavior patterns, providing information about their relative importance (proportion), which helps in road policy decision making in terms of development of prevention measures. The results help in decision making through a potentially better allocation of resources as carried out by road safety regulating offices such as the Spanish Traffic General Directorate (Dirección General de Tráfico, DGT).

Keywords

• Age
• driver behavior
• gender
• pattern recognition
• self-organizing maps (SOM)