U.S. State Energy Transition Dynamics: Renewable Adoption, Fossil Displacement, and Forecasting with Ensemble Machine Learning

The study investigates the dynamics of the United States’ energy transition in all 50 states (including the District of Columbia), over the period 1990 to 2024, using the United States Energy Information Administration’s State Energy Datasets. Using a combination of panel econometrics, clustering, and machine learning, this study aims to identify major patterns in the uptake of renewable energy sources in the United States, along with their economic effects. The preliminary results show significant cross-state heterogeneity and structural change in 2008 as the major turning point in the transition path. In addition, four major archetypes of states are identified, namely: early movers, progressive large-scale adopters, gradually transitioning states, and fossil-locked states, which are characterized by their high dependence on conventional sources of energy. Furthermore, a higher share of renewable energy reduces total energy expenditure, taking into account factors such as energy demand and prices. Based on the machine learning models, this study demonstrates a high degree of predictive accuracy (R² ≈ 0.77). Also, the interpretability analysis of the results demonstrates that lagged renewable share is the dominant predictor of future adoption. This implies a high degree of path dependence in which historical energy structures strongly constrain current transition dynamics. Thus, the U.S. energy transition is heterogeneous, cost-reducing, and structurally persistent, with inertia in renewable adoption representing a key barrier to accelerated decarbonization.