A Machine Learning Approach to Determining the CO2 Adsorption Capacity of Coconut Shell-Derived Activated Carbon

Abstract Coconut shell-derived activated carbon is widely used for the adsorption of gaseous contaminants including CO2 capture applications due to its availability, low costs, high surface area and tunable porous structure. However, determining the adsorption capacity of activated carbons through experimentation is challenging due to time constraints and the required equipment and experimental costs. This study aimed to develop a machine-learning model correlating the pore size distribution, pore volume, surface area, temperature, and pressure of activated carbons to their CO2 adsorption capacity. The Cochran model was used to determine the minimum number of data samples required to perform an unbiased representative analysis. Consequently, over 100 published coconut shell–derived activated carbon samples were collected from the open literature. A decision tree and linear regression model were developed to relate the pore volumes, pore diameter in different size intervals, surface area, temperature, and pressure to the maximum CO2 adsorption capacity. The model achieved good predictive accuracy with the decision tree regressor mean absolute error (MAE) of 4.49 on the test set. This data-driven machine learning model can be useful for predicting CO2 capacities based on synthesized pore structures and can become a useful tool for determining first estimates of CO2 adsorption capacity of coconut shell-derived activated carbon. The approach demonstrated here can be extended to model the adsorption of other gases on microporous carbons and utilized for software applications.