Evaluating the efficiency of spatial-geographical models for vehicle crash frequency estimation: A case study on the urban road network of hamadan province

• Spatial-geographical modeling was applied to estimate vehicle accident frequency. • Geographically Weighted Regression (GWR) and GWPR were utilized for spatial analysis. • The model incorporates environmental and geographical factors neglected in prior studies. • Hamadan’s urban road network was used as a case study to identify accident-prone segments. • Results show strong spatial autocorrelation between accidents and traffic volume variables. • Results support localized safety planning in underrepresented regions. • Spatial modeling improves predictions in data-scarce environments. This study evaluates the efficiency of spatial-geographical models in estimating vehicle crash frequency, focusing on the main urban roads network of Hamadan Province, Iran. Traditional models such as Generalized Linear Models (GLM) assume constant parameters across locations, while spatial models like Geographically Weighted Regression (GWR) and Geographically Weighted Poisson Regression (GWPR) capture spatial heterogeneity by allowing parameters to vary geographically. A total of 2,995 vehicle crashes (2,497 injurious and 496 fatal) between 2017 and 2019 were analyzed across 1,071 one-kilometer segments. The models were evaluated using multiple bandwidths in Kernel Density Estimation (KDE), where a 300-meter bandwidth yielded the best accuracy. GLM models using only AADT showed a parameter estimate of 0.75, while the full model including all predictors significantly improved the statistical fit (reduced AIC). GWPR models outperformed both GLM and GWR in terms of AIC, AICc, MAE, and RMSE. Spatial variation in the effect of variables such as traffic volume, road slope, and speed deviations revealed the importance of incorporating geographic heterogeneity in crash modeling. The findings suggest that localized modeling can lead to more accurate crash predictions and better-informed traffic safety interventions.