Design Guidelines for MOFs in CF4/NF3 Separation Based on Feature Selection with Stacking Ensemble Model

Efficient removal of carbon tetrafluoride (CF₄) from nitrogen trifluoride (NF₃) is essential for improving NF₃ purity in microelectronics manufacturing and mitigating the environmental impact of CF₄ due to its high global warming potential. Moreover, separating CF₄ from high-purity NF₃ presents a significant challenge. In this study, a comprehensive computational framework combining Grand Canonical Monte Carlo (GCMC) simulations and machine learning was employed to evaluate CF₄/NF₃ separation performance across 790 metal–organic frameworks (MOFs). A Stacking ensemble model integrating XGBoost, support vector regression (SVR), and artificial neural networks (ANN) was constructed, showing superior predictive performance over individual models. Shapley Additive Explanations (SHAP) analysis revealed that CF₄ adsorption is primarily governed by the Henry coefficient, adsorption enthalpy, and pore size parameters, with an optimal range of 8–12 Å for pore-limiting diameter. For selectivity, a novel descriptor─adsorption energy ratio (ratio_ADH)─was identified as the most effective predictor, exhibiting strong correlation with MOF selectivity performance. Based on these insights, rational design strategies were proposed, including the gradient placement of open metal sites, and construction of hierarchical pore architectures to simultaneously enhance CF₄ uptake and suppress NF₃ adsorption. This work provides a theoretical basis and data-driven guidance for the development of MOF-based adsorbents for CF₄/NF₃ separation, and lays the groundwork for future experimental validation and industrial application.