An analytics framework for predicting electric vehicle charging station demand using spatial and interpretive methods

The rapid adoption of electric vehicles (EVs) in Indonesia, supported by ambitious national targets, has outpaced the expansion of public charging infrastructure, particularly in Greater Jakarta. Ensuring the strategic siting of fast and ultra-fast electric vehicle charging stations (EVCS) is therefore essential to achieve high utilization, economic viability, and user satisfaction. However, planning in such early-stage EV markets is hampered by the scarcity of historical charging data, which limits the effectiveness of conventional demand-driven models. This study aims to develop a data-efficient framework for predicting EVCS demand under conditions of data scarcity. The approach integrates spatial interpolation, class balancing, and explainable machine learning to compensate for missing information and improve predictive reliability. Using a dataset of 137 sites with 21 spatial, socio-economic, and infrastructural features, the study evaluates the performance of multiple classifiers, with feature imputation conducted through Inverse Distance Weighting (IDW) and Ordinary Kriging. The results show that Nearest Neighbors on the SMOTE-balanced dataset—achieved an accuracy of 0.77 with imputed features, closely comparable to benchmark studies that relied on complete datasets (81.1%). SHAP analysis further highlights the most influential predictors of EVCS popularity, enhancing model transparency and interpretability. These findings demonstrate that spatial feature imputation, when combined with robust machine learning methods, provides a practical and replicable pathway for EVCS planning in data-scarce contexts. The framework offers actionable insights for policymakers and utilities, particularly in accelerating the deployment of fast and ultra-fast chargers in Indonesia’s metropolitan regions. • Develop an integrated analytics framework for predicting charging station demand. • Apply spatial interpolation to enrich features and improve predictive modeling. • Use interpretive methods to identify key drivers influencing site utilization. • Provide decision support for siting stations in data-scarce urban settings. • Demonstrate data-efficient modeling for infrastructure planning in emerging markets.