A machine learning framework for regional electricity–economic forecasting: Developing the U.S. Regional energy economic index (REEI) for equitable energy transition planning

Understanding how electricity systems reflect and shape regional economic transformation is increasingly important for energy transition planning. Existing studies largely rely on national-level aggregates or single-metric indicators, which obscure subnational heterogeneity in electricity and economic relationships and limit the ability to design regionally differentiated and equitable energy policies. To address this gap, this research develops a machine learning (ML) framework that links regional economic transformation to electricity consumption dynamics in the United States. Unlike prior national-level studies, this framework introduces a promising composite metric, the Regional Energy Economic Index (REEI), which uncovers the subnational disparities in the economic role of electricity through four system indicators: sales per customer, price volatility, revenue efficiency, and average price. We employ Gradient Boosting (GB), Random Forest (RF), and Extra Trees to harmonise datasets of the U.S. Energy Information Administration (EIA) and Residential Energy Consumption Survey (RECS 2020) to forecast household electricity demand with high accuracy (R2 = 0.791; CV R2 = 0.78). The study reveals significant regional differences in energy burdens, with a national average of 1.92%. Model interpretability via SHAP analysis classifies housing floor area and energy intensity as dominant drivers, while appliance and regional variables capture contextual heterogeneity. Non-parametric validations confirm statistically significant cross-regional discrepancies in both energy burden and consumption. The REEI also reveals three distinct transformation areas: industrial resilience, demand-volatility hotspots and constrained transition, highlighting geographically differentiated pathways in the national energy transition. This framework's robustness is further validated across PCA-, equal-weight-, and theory-weight-based constructions. These results, reported in a study that couples energy-economic indicators with explainable ML, highlight a scalable tool for planners and policymakers to identify high-energy burden regions, target infrastructure investments, and develop equitable decarbonization strategies. This study spots electricity data as a real-time proxy for regional economic vigor, enhancing data-driven planning and sustainability assessment within contemporary energy systems.