A probabilistic approach towards a crash risk assessment of urban segments

Abstract

This paper presents a probabilistic approach to measure the crash risk associated to an urban segment. This approach leads to a hotspot definition and identification using a probabilistic model defining the dependent variable as an indicator of a discrete choice. A binary choice model is used considering a binary dependent variable that differentiates a hotspot from a safe site set by the number of crashes per year per kilometre. The explanatory variables to set similar segments are based on average annual daily traffic, segment length, density of minor intersections. A threshold value for the number of crashes per kilometre is set to distinguish hotspots from safe sites. Based on this classification, a binary model is applied that allows the construction of an ordered site list using the probability of a site being a hotspot. A demonstration of the proposed methodology is provided using urban segment data from Porto, Portugal, covering a five-year period. The results of the binary model show a good fit. To evaluate and compare the probabilistic method with three usual hotspot identification methods described in the Highway Safety Manual, measures are used to test the performance of each method. Depending on the tests, actual data or simulated input, which are usually considered to set the "true" hotspots, were used. In general, the tests results indicate that the binary model performs better than the other three models. Nevertheless, it should be noted that the probabilistic approach provides an outcome that is quite different from the other methods, thus making difficult to ensure a linear comparison with the other methods. Overall, the study shows an alternative to hotspot identification using a risk measure in which the gains are the simplicity, the reliability, and the efficiency of model outcome.