Unveiling the adsorption behaviour of nitrogen-doped porous carbons (NDPCs) towards carbon dioxide capture using machine learning techniques

The urgent need to mitigate climate change has driven extensive research into effective carbon capture technologies. Among these, nitrogen-doped porous carbons (NDPCs) have emerged as a promising material for CO2 capture due to their large surface area, tunable porosity, and nitrogen functionalities. This paper investigates the adsorption behaviour of NDPCs towards CO2 using advanced machine learning (ML) techniques. These ML models are applied to different physicochemical properties of NDPCs to unveil the adsorption phenomenon of CO2 towards NDPCs. Gradient boosting decision trees (GBDT) performed best, with training and test R2 values of 0.99 and 0.88, respectively. Explainable machine learning (XML) has also been applied to find the intrinsic relation between NDPCs' properties and CO2 uptake. Combined factor partial dependence plots revealed the optimal range of the features for the CO2 uptake. Individual conditional expectation-partial dependence plots (ICE-PDPs) show the dependence of CO2 uptake on each instance of input features. To examine the impact of input features on CO2 uptake, Shapley Additive exPlanations (SHAP) have been employed. For nitrogen functionalities, the main impact came from pyridinic-N (N-6) and graphitic-N/quaternary-N (N-Q) compared to pyrolytic-N (N-5) and oxidized-N (N-X).