Spatial diagnostics and proxy-based demand modeling for EV charging infrastructure using machine learning

The current urban settings are experiencing increased pressures of the development of fair and effective electric vehicle (EV) charging facilities due to the unequal distribution of the population density and resources. This study will provide a geographically informed forecasting model based on ensemble machine learning algorithms, which are random forest regression (RFR) and extreme gradient boosting (XGBoost) to predict electric vehicle charging demand in the districts of Cairo. The model combines a combination of new engineered indicators that have been proposed in this paper like the charging accessibility score (CAS), station service radius (SSR), and connector diversity index (CDI) with more traditional variables, such as population density, station density, and power capacity. The initial random forest model was limited in its explanatory ability (R 2 = 0.30). The augmented XGBoost model significantly improved the predictive performance reaching a new R 2 of 0.76, reduced RMSE of 113,767–42,353 and mean absolute error (MAE) of 85,589–28,159. Spatial diagnostics showed that there are considerable differences in infrastructure with certain districts having more than 180,000 people per charging port, which shows the high-pressure areas that should be addressed at once. Both these indicators enhance geographic interpretability and predictability of the model. In comparison to conventional approaches to planning that are more traditional and thus more static, the proposed approach considers the variability in demand and identifies underserved regions, which can be a scalable and data-intensive tool to develop electric vehicle infrastructure in data-constrained urban environments in a more equitable and sustainable manner.