Data-driven explainable machine learning approaches for predicting hydrogen adsorption in porous crystalline materials

The efficient storage of hydrogen is crucial for its adoption as a sustainable energy carrier, addressing the growing need for clean energy sources. Porous crystals such as metal-organic frameworks (MOFs) exhibit exceptional potential among potential storage solutions due to their high surface areas, tunable pore structures, and strong adsorption capacities. However, optimizing these materials for practical hydrogen storage remains challenging, requiring advanced predictive techniques to accelerate material discovery and design. This study utilizes a comprehensive dataset comprising 14,544 experimental and simulated hydrogen adsorption samples to evaluate and develop advanced machine learning (ML) models. Five models, including Extreme Gradient Boosting (XGB), Gradient Boosting (GRB), Support Vector Regression (SVR), Multi-Layer Perceptron (MLP), and K-Nearest Neighbors (KNN), were applied to predict hydrogen uptake based on features including temperature, pressure, void fraction, specific surface area, and pore dimensions. XGB demonstrated superior performance with the lowest RMSE and highest R² values, ensuring accurate and generalizable predictions. Explainable AI techniques, particularly Shapley Additive Explanations (SHAP), were employed to interpret the models, identifying critical factors such as specific surface area and adsorption enthalpy as key drivers of hydrogen uptake. The results underscore the advantages of integrating ML models with experimental and computational datasets, offering a scalable framework for designing high-performance hydrogen storage materials. This research highlights the transformative potential of ML in hydrogen storage optimization, bridging the gap between computational modeling and material innovation. This study contributes significantly to developing sustainable energy technologies and global carbon neutrality initiatives by advancing predictive capabilities and offering interpretable insights.