<![CDATA[The strong nonlinear interaction between catalytic properties and operating conditions complicates accurate space time yield modeling in thermocatalytic carbon dioxide hydrogenation, especially when redundant descriptors are included. Although XGBoost is widely used for predictive tasks, the influence of feature redundancy on generalization and interpretability in carbon dioxide to methanol systems remains insufficiently examined. This study investigates the integration of Recursive Feature Elimination with XGBoost using 639 experimental observations derived from copper based catalysts. Reducing the feature set from fifteen to eight variables improves generalization performance, as indicated by lower prediction error and higher explained variance. The retained variables correspond to key catalytic and operational parameters, including reaction temperature, pressure, and copper content, aligning with established kinetic and mechanistic principles. These results show that eliminating redundant descriptors stabilizes cross validated performance and reduces training complexity without sacrificing predictive accuracy. The reduced model concentrates predictive weight on kinetically relevant variables, providing a clearer quantitative representation of the parameters that govern space time yield in carbon dioxide hydrogenation.]]>
Copyrights © 2026
