Fine particulate matter (PM2.5) pollution remains a critical global public health concern. Hence, this study investigates the determinants of PM2.5 concentration across G7 and EU nuclear-consuming countries from 1990 to 2021, focusing on the roles of energy supply mix, economic development, and technological innovation. Energy mix is operationalized as the disaggregated, source-specific elasticities of coal, oil, nuclear, and biofuels and their interactions with income and innovation, rather than a single composite diversification index. We combine panel regression methods with XGBoost machine learning and SHAP (Shapley Additive Explanations) to capture both linear and nonlinear relationships. The baseline model applies fixed-effects estimation with Driscoll–Kraay standard errors, while quantile regression and Feasible Generalized Least Squares (FGLS) are used for robustness. The results identify coal as the dominant driver of PM2.5 pollution, with a 1% increase in coal use raising levels by approximately 0.22%. Oil consumption also increases pollution, though to a lesser extent. In contrast, nuclear energy consistently reduces PM2.5 levels, while biofuels contribute to emission reductions only at higher adoption levels. Economic growth follows an inverted U-shaped pattern, supporting the Environmental Kuznets Curve hypothesis, a finding reinforced by SHAP analysis. Innovation activity significantly mitigates pollution across all models, and urbanization is associated with lower emissions in advanced economies. SHAP further reveals nonlinear dynamics, including S-shaped patterns for oil and inverted S-shaped relationships for biofuels, that traditional methods cannot capture. The findings point to four key policy priorities: accelerating coal phase-out, scaling biofuel adoption above critical thresholds, cautiously expanding nuclear capacity while monitoring diminishing returns at high penetration levels, and linking energy transitions to innovation investment for sustained air quality gains.