Resumen
Network safety ranking (or hotspot identification) is conducted in order to identify hazardous road locations and develop a priority list for application of effective safety treatments. In this process empirical Bayes approach, using accident prediction models, has been recommended. In order to be conducted efficiently, several questions need to be answered: for example which model function form should be adopted; which explanatory variables need to be collected and included in model; which time period is to be used. There is also a question whether each region requires its specific model or whether a transfer from elsewhere is feasible. These decision points motivated the presented study, which was conducted with data from national road network in two adjacent regions in the southeast of the Czech Republic (South Moravian region and Zlín region). For the purpose of network safety ranking, separate accident prediction models have been developed and applied in both regions. Several analysis steps were conducted in order to compare the model quality and performance, as well as both temporal transferability (between various periods) and spatial transferability (between the regions). The idea was to develop a transferable model, which would ease up on demands on data collection and modelling efforts and would thus increase effectiveness of the network safety ranking process. The presented study aims to assess the feasibility of developing such model.