An interpretable machine learning framework for SDG 7 electricity access forecasting and policy scenario analysis in sub-Saharan Africa

Abstract Universal electricity access by 2030 (Sustainable Development Goal 7, SDG 7) remains far off track in Sub-Saharan Africa (SSA), where approximately 565 million people lacked access in 2023. This study proposes an interpretable framework that integrates forecasting, explainability, and policy scenario analysis into a reproducible workflow, demonstrated on a World Development Indicators panel of 45 SSA countries (2001–2023; 1035 observations) with Ethiopia as the case study. Linear Regression (LR), Random Forest, XGBoost, and LightGBM are compared under leave-one-country-out and temporal-split validation. With path dependence represented through a lagged access feature, LR achieves the lowest error: a mean absolute error (MAE) of 1.93 percentage points across the 45 held-out countries (2.04 for Ethiopia) and 1.33 for Ethiopia in the temporal split. LR improves on a persistence baseline in 33 of 45 countries (median MAE improvement 15.9%). SHapley Additive exPlanations (SHAP) show that prior-year access dominates predictions, as expected under the autoregressive specification and consistent with infrastructure path dependence. SHAP informs the diagnostic assessment of the predictive structure, while multicollinearity screening, coefficient-sign verification, and scenario-sensitivity checks determine the selection of LR as the scenario model. Under Slow, Business-as-Usual, and Accelerated scenarios, Ethiopia is projected to reach 67.5–69.9% access by 2030, leaving a gap of 30–32 percentage points. Within the socioeconomic scenarios represented, macroeconomic improvements alone are unlikely to close this gap; structural interventions such as grid extension, lower connection costs, and off-grid solar deployment are needed. The framework is open and potentially transferable, pending validation elsewhere.