Development of the\nCO<sub>2</sub> Adsorption Model\non Porous Adsorbent Materials Using Machine Learning Algorithms

Porous adsorbents have common characteristics, such as\nhigh porosity\nand a large specific surface area. These characteristics, attributed\nto the internal structure of the material, significantly affect their\nadsorption performance. In this research study, we created a data\nset and collected data points from porous adsorbents (2789) from 21\npublished papers, including carbon-based, porous polymers, metal–organic\nframeworks (MOFs), and zeolites, to understand their characteristics\nfor CO<sub>2</sub> adsorption. Different machine learning (ML) algorithms,\nsuch as NN, MLP-GWO, XGBoost, RF, DT, and SVM, have been applied to\ndisplay the CO<sub>2</sub> adsorption performance as a function of\ncharacteristics and adsorption isotherm parameters. XGBoost was selected\nas the best ML algorithm due to its highest accuracy (<i>R</i><sup>2</sup> = 0.9980; MSE = 0.0001). The predicted results revealed\nthat the adsorption pressure parameter is the most effective in all\nof the mentioned porous adsorbents. With regard to materials type,\nwhile carbon-based materials require higher pressures for a more effective\nCO<sub>2</sub> adsorption, MOFs exhibit a higher potential for adsorbing\nCO<sub>2</sub> under lower pressure conditions. The study also revealed\nthat carbon-based adsorbents, zeolites, and porous polymers with smaller\npore diameters demonstrate a high level of CO<sub>2</sub> uptake.\nIn contrast, the adsorption performance of MOFs does not show a consistent\ntrend with respect to pore sizes. Also, in all adsorbents, the effect\nof a pore size smaller than 1 nm on more CO<sub>2</sub> adsorption\nwas evident.