Enhancing Predictive Accuracy of Hydrogen Storage Performance in Metal-Organic Frameworks Through Stacking Learning
Key Ideas
- 12 machine learning models were evaluated, with FusionStackBoost showing significant improvements in accuracy and overfitting reduction.
- Crystallographic features like Density, GSA, VSA, VF, PV, LCD, and PLD were crucial in predicting hydrogen storage in MOFs.
- ML models like XGBoost, LightGBM, and CatBoost were effective in predicting Useable Gravimetric and Volumetric Hydrogen Capacity.
- The study emphasizes the importance of multi-model collaboration and meta-learning in optimizing hydrogen storage performance in MOFs.
Metal-organic frameworks (MOFs) have emerged as promising materials for hydrogen storage due to their adjustable pore structures, but predicting their performance accurately has been challenging due to complex relationships. This study utilized the HyMARC database to evaluate 12 machine learning models for predicting hydrogen storage in MOFs. Models like XGBoost and LightGBM stood out for their accuracy. A novel FusionStackBoost model was developed using meta-learning, significantly reducing overfitting and enhancing prediction accuracy. Crystallographic features like Density and PV were identified as key in predicting hydrogen storage. The study highlighted the importance of multi-model collaboration and meta-learning in achieving precise predictions. This research not only improves accuracy in predicting hydrogen storage but also provides insights for optimizing MOFs for efficient hydrogen storage.