Using a k-means clustering algorithm to examine patterns of vehicle crashes in before-after analysis

Abstract

The study aims to develop a support procedure to estimate the efficacy of infrastructural interventions to improve road safety. The study was carried out on a 110 km stretch of the A3 highway, in southern Italy. Data from a huge sample concerning traffic, geometry and accidents for two periods of the same duration were compared, for which cluster analysis, and in particular, the "hard c means" binary partition algorithm was employed. Using cluster analysis, all the accidents with strong similarities were aggregated. Then for each cluster, the "cluster representative" accident was identified, to find the average among the various characteristics (geometrical, environmental, accident-related). A "hazard index" was also created for each cluster, whereby it was possible to establish the danger level for each "cluster". Using this information, an accident prediction model using a multi-variate analysis was produced. This model was used as a support for decision-making on infrastructures and to simulate situations to which the Before-After technique could be applied.