Advanced machine learning framework for accurate and interpretable CO ₂ capture predictions in porous carbon materials

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:This ML stacking ensemble method provides a reliable tool for predicting CO2 uptake in porous carbons, aiding material screening.

With the pressing environmental conservation agendas, effective CO2 capture methods are sought after. To address this issue, a study was conducted to analyse the existing literature and gain insights into the mechanisms of CO2 adsorption. In this study, a dataset of 311 biomass-derived porous carbon samples that included both intrinsic material properties (pore structure, elemental ratios) and extrinsic process-related parameters (pressure, temperature) is compiled and analysed. Various machine learning approaches, such as gradient-boost decision trees (GBDT), XGBoost, and LightGBM, were applied and compared their performance with standard regression metrics. Further, to improve the prediction efficiency, two Ridge regression meta-learner-based stacked ensembles, such as Random Forest (RF) + support vector regression (SVR), and RF + GBDT, are applied and evaluated. Both stacking approaches improved prediction reliability; however, the RF+SVR model performed much better than the RF+GB model. Further, the interpretation of models through feature importance analysis and partial-dependence analysis showed important features, such as pressure, surface area, that provide more insights to understand the inherent characteristics of CO2 adsorption. Thus, application of ML-based ensemble modelling confirms that this approach can be relied upon with high confidence to predict the CO2 uptake of new carbon materials and provides insights into the material development.

• openalex https://doi.org/10.1049/icp.2026.1032first seen 2026-07-09 05:17:07