Accelerating CO2 uptake modeling in carbon-based adsorbents through machine learning

Abstract

Climate change is driven by numerous factors but one of the primary drivers includes the rapid increase in atmospheric carbon dioxide (CO 2 ) concentration. This necessitates the need for the development of efficient carbon capture technologies. CO 2 adsorption through porous carbon-based materials has emerged as an innovative solution due to its low-cost and efficiency. Among the available porous carbon-based materials, carbon-based monoliths are a class of complex materials widely used for carbon capture. In this work, eight different machine learning models were employed on a dataset of 2215 data points collected from peer-reviewed published literature to predict the CO 2 adsorption capacity of carbon-based monoliths. For modelling, various parameters such as the BET specific surface area, temperature (T), pressure (P), micropore volume, mesopore volume, and elemental compositions were taken into consideration. Further, statistical parameters such as root mean square error (RMSE) and the coefficient of determination (R 2 ) were evaluated to assess the best-performing model. Among the models evaluated, the Gradient Boosting Decision Tree outperformed other models achieving an R 2 of 0.9958 and RMSE of 0.0773, indicating its superior robustness and predictive ability in capturing complex relationships that influence CO 2 adsorption. Conversely, the AdaBoost model exhibited the least effective prediction with an R 2 of 0.9398 and RMSE of 0.2925. Moreover, SHAP feature analysis revealed that T and P were the most influential parameters for CO 2 adsorption, while nitrogen (wt%) was the most critical in terms of material composition. The developed models can serve as a pre-screening tool to accelerate material discovery for CO 2 adsorption while reducing experimental costs and optimizing the design of carbon capture systems.