| Resumo: | A theoretical definition of a hotspot is any location that has a higher expected number of accidents than other similar locations as a result of local risk factors present at the location. This study presents an alternative approach to research regarding hot spot definition and identification based on a probabilistic model that defines the dependent variable as an indicator of a discrete choice. A binary choice model was used considering a binary dependent variable that differentiates a hot spot (category 1) from a safe (category 0) site set by the number of accidents per kilometer. To define these two categories, it is necessary to use a strategy that selects a hot spot as accurately as possible. Based on a threshold strategy, various hypotheses were analyzed to obtain the most appropriate value to balance sensitivity and specificity criteria (epidemiological criteria). To apply this approach, risk factors including traffic volume, the number of minor intersections per kilometer, the road function classification and land use from an urban segment data set collected over a five-year period from Porto, Portugal, were used. The probabilities were estimated by the binary choice model, and a performance evaluation was then applied. In addition, considering the probabilistic nature of this approach, it is also possible to define four classes to classify a site in terms of safety using the uncertainty in a site being a hot spot, setting for each class a range of probability values. Furthermore, a comparative analysis was considered to test the performance of the qualitative response (QR) method relative to two commonly applied methods - accident frequency (AF) and empirical Bayes (EB). Considering the probabilistic nature of this approach, two probability threshold values were determined based on the "correct proportion" of sensitivity and specificity. These two values, along with the 0.5 probability value that the model defines to separate a hot spot from a non-hot spot, were used to define four classes. Using these classes, priorities and types of engineering intervention can be determined and thus, site interventions can be efficiently managed. In addition, the QR approach was compared with the AF and EB, against four robust and informative quantitative evaluation criteria tests. These tests showed that the QR method performs better than the two other methods. In this paper, a strategy to more accurately classify hot spots from a set of sites is proposed using four classes that may define priorities and types of engineering intervention, allowing for the efficient management of site intervention. Above all, this analysis proves that the application of a probabilistic model is an alternative approach to hot spot research.
|
|---|