MACHINE LEARNING-DRIVEN THERMODYNAMIC PREDICTION OF HYDROGEN STORAGE PERFORMANCE IN METAL HYDRIDES FOR SUSTAINABLE ENERGY APPLICATIONS

Abstract

Efficient hydrogen storage remains a major challenge in the development of sustainable energy systems. Metal hydrides have emerged as promising solid-state hydrogen storage materials because of their favorable thermodynamic stability and reversible hydrogen absorption behavior. This study aimed to predict hydrogen storage performance in metal hydrides using machine learning techniques and to identify the variables most strongly influencing Hydrogen Weight Percent. A quantitative computational framework was implemented using secondary thermodynamic data containing 772 observations and 13 variables. Data preprocessing included missing-value treatment, categorical encoding, and feature scaling prior to model development. Three regression algorithms, namely Linear Regression, Random Forest, and Gradient Boosting, were evaluated using R-squared, mean absolute error, root mean square error, and five-fold cross-validation. The results showed that Gradient Boosting achieved the highest predictive performance with an R squared value of 0.9088, a mean absolute error of 0.1182, and a root mean square error of 0.2194. Random Forest also demonstrated strong predictive capability, whereas Linear Regression produced comparatively lower accuracy. Cross-validation results confirmed the robustness and generalization capability of the ensemble learning models. Feature importance analysis revealed that HtoM was the most influential predictor, contributing 57.77% of the total model importance, followed by Material Class, Temperature ^oC, and Pressure Atmospheres Absolute. The findings demonstrate that machine learning provides an efficient computational approach for hydrogen storage prediction and supports accelerated screening of metal hydride materials for sustainable energy applications.