Application of Geographically Weighted Regression Technique in Spatial Analysis of Fatal and Injury Crashes

Abstract

Generalized linear models (GLMs) are the most widely used models utilized in crash prediction studies. These models illustrate the relationships between the dependent and explanatory variables by estimating fixed global estimates. Since crash occurrences are often spatially heterogeneous and are affected by many spatial variables, the existence of spatial correlation in the data is examined by means of calculating Moran’s I measures for dependent and explanatory variables. The results indicate the necessity of considering spatial correlation when developing crash prediction models. The main objective of this research is to develop different zonal crash prediction models (ZCPMs) within the geographically weighted generalized linear model (GWGLM) framework in order to explore the spatial variations in association between number of injury crashes (NOICs) (including fatal, severely, and slightly injured crashes) and other explanatory variables. Different exposure, network, and sociodemographic variables of 2,200 traffic analysis zones (TAZs) are considered as predictors of crashes in the study area, Flanders, Belgium. To this end, an activity-based transportation model framework is applied to produce exposure measurements while the network and sociodemographic variables are collected from other sources. Crash data used in this study consist of recorded crashes between 2004 and 2007. The performances of developed GWGLMs are compared with their corresponding GLMs. The results show that GWGLMs outperform GLMs; this is due to the capability of GWGLMs in capturing the spatial heterogeneity of crashes.