Avaliação da dependência espacial na modelagem do desempenho da segurança viária em zonas de tráfego

Abstract

A common technique used in the modeling process of the Road Safety Performance at the planning level is the Generalized Linear Models (GLM) procedure with the assumption of negative binomial error distribution. A main limitation of this technique, which is the no consideration of spatial effects, has been overcome by the use of local spatial models such as the Geographically Weighted Poisson Regression (GWPR). This work aims to present a comparative analysis between non spatial global and spatial local accident prediction models focused to estimate the safety performance of traffic accident zones of Fortaleza city. Models were calibrated to thedependent variable total accidents and accidents with victims and the results showed that GWPR models performed better than GLM on measures of adjustment and the reduction of residual spatial autocorrelation, being able to capture the spatial heterogeneity in the frequency of accidents.