A Geo-Spatial Multi-Table Stacked Ensemble Model for Forecasting Electric Vehicle Charging Infrastructure Readiness and Expansion Priority

Abstract The accelerating adoption of Electric Vehicles (EVs) has placed unprecedented pressureon global charging infrastructure, exposing significant disparities between EVpenetration and the spatial distribution of charging stations. Traditional planningframeworks rely on single-source datasets and simple statistical models, whichinadequately capture the geo-spatial, technological and policy-driven complexities of EV ecosystems. This paper proposes a Geo-Spatial Multi-Table Stacked Ensemble (GSMT-SE) model that integrates five heterogeneous datasets—station-level operational records, machine-learning ready features, country-level summaries, global statistics and EV modelspecifications—into a unified analytical framework. The proposed model fuses K-Means based geo-spatial clustering with engineered features such as charging density, fast-charging indicators and accessibility scores, and stacks Random Forest, XGBoost and a Multi-Layer Perceptron base learner under a meta-learner to perform multi-taskprediction of charging readiness and expansion priority. Experimental results on a globalEV charging dataset demonstrate that the GSMT-SE model achieves an accuracy of0.951, F1-score of 0.950 and ROC-AUC of 0.982, outperforming baseline learnersincluding Logistic Regression, Decision Tree, Random Forest and XGBoost. Featureimportance analysis confirms that power capacity, country station density and spatialcluster membership are the dominant predictors of infrastructure readiness. The proposedframework provides actionable, data-driven insights for policy makers, urban plannersand charging-network operators seeking to prioritise EV infrastructure investment insupport of sustainable mobility goals.