At a European level, road accident fatalities on two-lane rural roads constitute about 60% of the total number of fatalities caused by road accidents. This paper presents a comprehensive procedure to define an Accident Prediction Model (APM) for two lane road sections based on road features related to the safety performance of the stretch. Firstly, significant variables related to the exposure, geometry, consistency and context factors were identified for the segmentation of highways into homogeneous sections.All the basic data were obtained from a low cost surveys purposely set up. In particular, differential cinematic Global Positioning System (GPS) surveys were used to define the horizontal alignment and Road Safety Inspections (RSIs) were carried out in order to quantify the other road characteristics related to safety. The latter part of the paper focuses on the calibration of APMs able to estimate the expected number of accidents on homogeneous sections of different lengths that can be characterized by constantvalues of the explanatory variables previously defined. Several APMs werecalibrated using the Generalized Linear Model approach (GLIM). Finally, considering the statistical significance of the parameters related to exposure, geometry, consistency and context factors, and the model's goodness of fit measures, 19 models were ranked and three APMs were selected as bestmodels. The first one is a base model, function of length and traffic only, which can be used along with an Empirical Bayesian calculation to conduct network screening analysis. The other two models represent the best statistical fit with significant variables related to exposure, geometry, consistency and context factors.
Abstract