Development of planning level transportation safety tools using Geographically Weighted Poisson Regression

Area of Science:

• Transportation Science
• Spatial Statistics
• Traffic Engineering

Background:

• Generalized Linear Modeling (GLM) is common for collision prediction but assumes stationary relationships.
• This assumption may obscure spatial variations in collision frequency influenced by land use, demographics, and traffic volume.
• The accuracy of GLM for explaining collision-occurrence relationships is potentially limited by unaddressed spatial factors.

Purpose of the Study:

• To investigate spatial variations in the relationship between zonal collisions and transportation planning predictors.
• To compare the accuracy of Geographically Weighted Poisson Regression (GWPR) against conventional Generalized Linear Models (GLM).

Main Methods:

• Application of Geographically Weighted Poisson Regression (GWPR) modeling technique.
• Comparison of GWPR model performance with Generalized Linear Models (GLM).

Main Results:

• Geographically Weighted Poisson Regression (GWPR) models effectively capture spatially dependent relationships in collision data.
• GWPR models generally demonstrate superior accuracy compared to traditional Generalized Linear Models (GLM).

Conclusions:

• Geographically Weighted Poisson Regression (GWPR) offers a more robust approach for collision prediction by accounting for spatial heterogeneity.
• The findings highlight the importance of spatial analysis in transportation planning and traffic safety.