Selection of MOF Functional Materials for CO₂ Capturing Using an Integrated Isotherm and Machine Learning Models.

This work presents an integrated experimental and machine-learning (ML) framework for evaluating Zr-based trimetallic metal–organic frameworks (MOFs) for CO₂ capture. Three MOFs, namely MOF-1 (Zr–Fe–Cu–BTC), MOF-2 (Zr–Ni–Mn–BDC), and MOF-3 (Zr–Cr–Zn–BTC), were systematically characterized to correlate structural and textural properties with CO₂ adsorption performance. Among them, MOF-2 exhibited the highest BET surface area (642.4 m² g⁻¹) and CO₂ uptake (0.764 mmol g⁻¹ at 298 K and 100 kPa). An extensive literature dataset containing 8,206 MOF entries was subsequently used to develop ML regression models, with support vector regression achieving R² > 0.999 for the evaluated isotherm data. SHAP, PCA, and t-SNE analyses were employed to identify influential descriptors and visualize structural relationships within the dataset. Importantly, transferability tests using the experimentally synthesized MOFs revealed limitations in applying literature-trained models to low-porosity and chemically distinctive frameworks, highlighting that conventional geometric descriptors alone may not adequately represent CO₂–framework interactions. By combining experimental characterization, classical adsorption modelling, ML prediction, and explainable analysis, this work provides a data-driven framework for understanding MOF–CO₂ adsorption relationships and supporting the development of more transferable models for next-generation CO₂ capture materials.