Analyzing Public Electric Vehicle Charging Distribution and Urban Socio-Spatial Characteristics Using Geographically Weighted Machine Learning

To understand the equitable distribution of public electric vehicle charging infrastructure (EVCI), previous studies have used geographically weighted regression (GWR) to evaluate the spatial heterogeneity between EVCI and urban socio-spatial characteristics. However, GWR is susceptible to outliers and multicollinearity and overlooks the nonlinear relationships inherent in complex urban dynamics. This study advances EVCI distribution evaluation by incorporating two geographically weighted machine learning approaches, random forest (GW-RF) and extreme gradient boosting (GW-XGB), while using GWR for comparison. Focusing on nonlinear spatial relationships, we examined the EVCI distribution in King County, Washington, considering 20 socio-spatial factors. We interpreted the within-sample feature importance scores by combining the two machine learning approaches. Compared with GWR, we found that population density and multi-unit housing proportion are the two key factors related to EVCI distribution. The results indicate that future EVCI planning should prioritize residential charging needs to support broader transportation electrification goals across diverse communities.