Machine learning prediction of hydrogen sorption properties of zirconium-titanium based C14 Laves phase AB 2 alloy from chemical composition

Abstract Machine learning (ML) models were developed to predict the hydrogen sorp-tion properties of Zr–Ti-based C14 Laves phase AB 2 alloys—unit cell volume (V14), enthalpy of hydrogen absorption (∆H), entropy of hydrogen absorption (∆S), and equilibrium plateau pressure (P eq)—from chemical composition and calculated physical properties. A curated dataset of 244 alloys was compiled from the literature, with two feature categories compared: element atomic percent and properties-based variables (molar mass, bulk density, molar volume, atomic radius ratio, bulk modulus, electron density, B/A ratio). Five ML algorithms (linear regression, random forest, ridge regression, KNN, XGBoost) were trained with 80/20 train–test split, 5-fold cross-validation, and forward feature selection. Element atomic percent features outperformed properties-based features for all target variables. V14 was predicted with the highest accuracy (R 2 up to 0.955 on the test set), while P eq showed moderate accuracy (R 2 up to 0.741) and a notable linear correlation with V14. Prediction of ∆H and ∆S yielded moderate and low accuracies, respectively, highlighting opportunities for more advanced feature engineering. These results demonstrate the potential of ML to accelerate the design of AB 2 hydrogen storage alloys.