Data-driven prediction of mechanical properties of Ti-based biomedical alloys and their high-entropy extensions using tree-based ensemble learning

Titanium-based Bio High-Entropy Alloys (Bio-HEAs) are gaining attention for their excellent mechanical properties and biocompatibility, making them ideal for biomedical use. In this study, a tree-based ensemble machine learning framework is developed to predict four key mechanical properties including Young’s Modulus (YM), Yield Strength (YS), Ultimate Tensile Strength (UTS), and Hardness (HV) of Ti-based Bio-HEAs. A dataset of 514 experimentally reported alloys was collected from literature spanning over past 20 years. Missing data was handled with iterative imputation, and the dataset was cleaned and normalized. Thermo-dynamic and semi-empirical parameters were computed using the High Entropy Alloy Predicting Software (HEAPS). Multiple tree-based ensemble learning algorithms, including Gradient Boosting, Extra Trees, Random Forest, XGBoost and CatBoost Regression were trained using an 80:20 train-test split. To further improve prediction accuracy, a voting ensemble model was evaluated. A combination of filter, wrapper, and embedded feature selection methods - namely SelectKBest (f regression), Recursive Feature Elimination (RFE), and Random Forest - based feature importance, were applied to identify the most relevant features, and model interpretability was improved by Partial Dependence Plots (PDPs). Optimized model attained strong predictive performance with R 2 scores of 0.94 (YM), 0.73 (YS), 0.96 (UTS), and 0.82 (HV). The study highlights the potential of data-driven methods in design and screening of biomedical alloys through reducing need for trial-and-error experimentation.