Interpretable machine learning for predicting electric vehicle adoption in a spatially heterogeneous landscape

This work presents a machine learning framework to predict the adoption of Light-Duty Electric Vehicles (EVs) across Southern California at the Census Block Group (CBG) level. Using Explainable Boosting Machine (EBM), we model spatial, infrastructural, and sociodemographic drivers of adoption, incorporating indices of disadvantage, pollution burden, health risk, grid readiness, and spatial diffusion. We analyze data for 14,979 CBGs with 66 variables from fleet, energy, and census sources. Results indicate EV adoption is shaped by more than infrastructure: urbanicity, auto ownership, work access, and disadvantage also influence uptake. Infrastructure shows weaker associations with adoption in disadvantaged areas. The EBM model achieves strong performance (R2=0.73), performing competitively with Random Forest, LightGBM, and XGBoost while remaining interpretable. Underserved regions are found to have latent demand, thereby supporting the development of targeted charging deployment aligned with Justice40 equity priorities. The framework is transferable to local, regional, and state-level EV planning.