Hydrogen storage capacity in metal-organic frameworks: Towards elevating predictions through ensemble learning with a comprehensive preprocessed dataset

Abstract

This work predicts hydrogen storage in metal-organic frameworks (MOFs) utilizing ensemble learning models on preprocessed data. Features including Brunauer–Emmett–Teller (BET) and Langmuir surface area, pore volume, pressure, temperature, and isosteric enthalpy of adsorption are used to predict excess gravimetric storage. A tailored data preprocessing approach has been implemented on the raw experimental data to enhance the dataset's quality. Model evaluation across Random Forest, eXtreme Gradient Boost (XGBoost), Categorical Boosting (CatBoost), and Light Gradient Boosting Machine (LightGBM) reveals CatBoost's as the most effective model (R² = 0.994, Root Mean Square Error = 0.019). Leverage method validation affirms the domain's applicability, while sensitivity analysis illuminates feature significance and wise feature selection. This study advances hydrogen storage by providing a robust predictive framework that bridges theoretical insights with practical applications. These proposed predictive models can be used for finding the best candidate, optimizing operational conditions, and for subsequent applications such as hybrid modeling purposes.