Predictive modeling and feature attribution of CO₂ adsorption on LDH-derived materials using machine learning approach

Decarbonization has become a global necessity to combat climate change and build a sustainable future for generations to come. Solid porous materials have been identified as promising materials for CO₂ adsorption, which is an essential measure towards climate action. Layered double hydroxides (LDH) are one of the efficient materials used for carbon capture applications. We report the application of six machine learning models, namely Random Forest (RF), gradient boosting decision tree (GBDT), extreme gradient boosting (XGBoost), Light gradient boosting machine (LightGBM), categorical boosting (CatBoost), and Adaptive boosting (AdaBoost) to predict CO₂ adsorption capacity on LDH-derived materials. The findings revealed that CatBoost is most suitable with R² of 0.99 for training and 0.87 for test, and RMSE of 0.184 compared to the other 5 algorithms. AdaBoost, XGBoost, GBDT, LightGBM, and RF performed in an acceptable range. Two-factor dependence plots offered an explicit view of how the interaction between two input features influences the model’s prediction. SHapley Additive exPlanations (SHAP), an advanced interpretability method, was applied to the CatBoost model to investigate CO₂ adsorption behaviour. SHAP analysis exposed strong relationships between input features and outputs, provided clear feature importance, and quantified individual feature importance on CO₂ adsorption behaviour. 3D combined feature importance results reveal non-linear trends, including dual regimes of favourable interlayer spacing and the synergistic effect of surface porosity with lattice structure. Moderate calcination of the material emerges as the optimal synthesis window, which results in balanced porosity, structural stability, and basic site density. These findings provide mechanistic insight and practical guidelines for tailoring LDH sorbents to maximise CO₂ capture performance. Overall, this study not only delivers data-driven insights for CO₂ capture over LDH-derived materials but also establishes a reliable predictive framework that eliminates the need for experiments and creates new possibilities for adsorption research using large datasets and thus proving an effective tool towards decarbonization.