An explainable machine learning framework for predicting energy security resilience across G7 economies

Ensuring energy security amid geopolitical disruptions and accelerating net-zero transitions has emerged as a defining policy challenge for G7 economies. Yet reliable, forward-looking, and country-specific quantification of energy security resilience (ESR), operationalized here as energy mix diversification through the complement of the Herfindahl–Hirschman Index (1−HHI), remains limited. Existing approaches rely predominantly on static index frameworks or pooled econometric models that overlook national nonlinearities, fail to account for uncertainty, and offer insufficient interpretability for policymaking. This study addresses these gaps through a country-specific explainable machine learning pipeline integrating three innovations: exhaustive multi-criteria feature selection, including Pearson correlation, mutual information, and Random Forest importance, tailored to each G7 nation; comparative multi-algorithm modelling, including Ridge, Lasso, Huber, and Random Forest, with 500-iteration bootstrap 95% confidence interval estimation; and SHAP-driven model transparency to decompose prediction variance into individual associational drivers. Random Forest emerged as the optimal model for five of seven countries, yielding out-of-fold R 2 values between 0.80 and 0.97, while Lasso and Ridge best served Canada and the United States, respectively. Permutation significance tests confirm that results exceed chance performance ( p < 0.05) for five countries. Energy efficiency and emission intensity consistently rank as the dominant long-run associational drivers of energy mix diversification, with purchasing power and trade openness exerting significant but context-dependent effects. Forward projections to 2040 reveal divergent national trajectories: Italy ( + 6.6%) and Japan ( + 6.0%) demonstrate the strongest resilience gains, whereas Canada ( − 1.0%) and the United Kingdom ( − 1.4%) face structural decline. These findings provide policymakers with interpretable, uncertainty-bounded evidence to prioritize investments in clean technology, grid modernization, and diversified supply chains.