An Explainable Stacked Machine Learning Approach for Toluene Capture in Metal‐Organic Frameworks: From Predictive Modeling to Interactive Web Platform

ABSTRACT Metal‐organic frameworks (MOFs) exhibit significant potential for the adsorption of volatile organic compounds (VOCs) due to their tunable pore structures and high specific surface areas. However, identifying high‐performing MOFs within the vast structural space remains challenging, primarily due to unclear structure–performance relationships. Moreover, existing studies often overlook realistic adsorption scenarios that involve coexisting atmospheric components such as O 2 , N 2 , and water vapor, and rarely address capacity–selectivity trade‐offs or conducted systematic comparisons of model performance. Herein, we developed a data‐driven machine learning framework integrating multi‐model comparisons, stacking ensemble modeling, and interpretability analyses for predicting the adsorption performance of MOFs for airborne toluene with high accuracy. The stacking model, comprising eight complementary base models and a multilayer perceptron (MLP) as the meta‐learner, demonstrated an enhanced capability to capture complex nonlinear relationships between descriptors and performance, achieving superior predictive accuracy ( R 2 = 0.922, RMSE = 0.186) compared to the best‐performing individual model, CatBoost ( R 2 = 0.890, RMSE = 0.326). Furthermore, by incorporating SHAP, PDP, and feature interaction analyses, this study elucidated the synergistic regulatory mechanisms associated with key structural descriptors. Statistical analysis further revealed that the structural parameters of high‐performing MOFs exhibited significant convergence, with metal centers such as Cu and their open metal sites (OMS) quantitatively identified as critical performance‐enhancing factors. Finally, the stacking model was successfully deployed as an interactive web platform that enables real‐time prediction and visual interpretability of MOF performance, serving as a practical tool for the efficient screening of MOF candidates for airborne toluene adsorption.