ABSTRACT Catalytic hydrogenation of CO 2 to methanol is a promising route for carbon recycling and sustainable fuel production, but performance depends on coupled effects of catalyst properties and operating conditions. Here, we present an interpretable machine‐learning framework integrating prediction, interpretation, and multi‐objective optimization to accelerate design. Using 1982 experimental entries, we benchmarked several regression models and identified XGBoost as the best surrogate, with R 2 ≈ 0.83. SHapley Additive exPlanations (SHAP) analysis showed that methanol space time yield (STY) is governed mainly by gas hourly space velocity (GHSV), temperature, and pressure, with strong nonlinear interactions beyond single‐factor trends. GHSV × temperature and GHSV × pressure were the most important interaction pairs. Design maps and Non‐dominated Sorting Genetic Algorithm II (NSGA‐II) Pareto fronts identified a balanced operating window that maximizes STY while limiting thermal and pressure severity. This framework can be extended to other catalytic systems with strong multivariable coupling.
Asif et al. (Wed,) studied this question.