Machine Learning-Guided Discovery of Efficient Metal-Organic Frameworks for Volatile Organic Compound Removal: A Case Study on CCl4.

The adsorption of carbon tetrachloride (CTC) in MOFs was investigated by integrating GCMC simulations at 298.15 K and 10 kPa with machine learning (ML). The structural properties, including the largest cavity diameter, pore limiting diameter, and accessible surface area, were used as features in the ML models. Four algorithms, including Random Forest, LightGBM, XGBoost, and CatBoost, were employed to predict the CTC adsorptions. CatBoost demonstrated superior accuracy, achieving an R2 of 0.859 and an RMSE of 0.049. K-Means clustering identified MOF clusters with high average adsorptions and highlighted the prevalence of metals such as Zn, Co, and Cd. The key features influencing adsorption included heat of adsorption, specific accessible volume, gravimetric surface area, volumetric surface area, and specific accessible volume. Mann-Whitney U analysis of the top performing MOFs revealed that Void Fraction, gravimetric surface area, and specific accessible volume are the most important features for the adsorption of CTC in MOFs.