Machine learning boosted optimal parameters identification for higher alcohols synthesis over CuZn-based catalysts with experimental validation

Higher alcohols synthesis from syngas presents a promising route to reduce dependence on crude oil and also promote clean coal utilization. However, trial-and-error experimentation and laborious theoretical calculations fail to satisfy the demands of rational catalyst design. Here we developed a machine learning framework utilizing a SMOGN-preprocessed database of 790 CuZn-based catalyst experimental data to predict CO conversion and higher alcohols selectivity via an optimized XGBoost model (R 2 = 0.915 and 0.837) which were filtered from four different ML algorithms. Feature importance analysis uncovered the Cu particle size (35–57 nm) as one of the most pivotal descriptors for both reactant conversion and product selectivity. The SHAP analysis further revealed that a critical Cu 0 /Cu + ratio (1.8–2.0) threshold dictates a shift from negative to positive influence of larger Cu particles on catalytic performance. Increasing Cu/Zn ratio (3.5–5.0) correlated positively with higher alcohols selectivity. A Pareto frontier analysis of 29,946 XGBoost-predicted virtual data pointed explicitly quantified the inherent trade-off between CO conversion and C 2+ alcohols selectivity, with the ML-predicted optimal descriptor window clustering near the upper Pareto boundary. Moreover, an optimal structure-activity relationship interval was established via leveraging grid parameter combinations analysis. Experimental validation using three catalysts spanning sub-optimal and ML-predicted optimal descriptor spaces confirmed the directional predictions of the framework: catalysts within the optimal Cu/Zn window (3.5–5.0) exhibited significantly enhanced total alcohol selectivity, and exclusively displayed bimodal Anderson-Schulz-Flory distributions consistent with the dual chain-growth pathway predicted by SHAP analysis. The work offers a reliable strategy and robust data support for accelerating the rational design of CuZn-based catalysts for higher alcohols synthesis. • The machine learning framework was developed to optimize key parameters for CuZn-based catalyst discovery. • The XGBoost model excelled in predicting C 2+ alcohols selectivity (R 2 = 0.915) and CO conversion (R 2 = 0.837) with highest accuracy. • Optimal Cu 0 /Cu + ratio (1.8–2.0), Cu/Zn ratio (3.5–5.0), and Cu particle size (35–57 nm) maximized target performance. • Experimental validation confirmed the model's accuracy in guiding higher alcohols synthesis.