Abstract:TiZr-based refractory high-entropy alloys (RHEAs), known for their high hardness, compressive strength, and thermal phase stability, have garnered attention due to their potential application in armor-piercing warheads. This study introduced a machine learning (ML)-assisted approach to alloy design, aiming to uncover the complex relationships between composition and performance and to improve design efficiency. To address the critical requirement for hardness in armor-piercing applications, a 15-dimensional feature dataset was constructed from 157 experimental hardness data points, incorporating component molar fractions and five key descriptors. Eight ML models, including random forest, K-nearest neighbors, and support vector machines were trained, and XGBoost was identified as the most accurate through hyperparameter tuning via grid search and cross-validation. The SHAP (Shapley additive explanations) framework was applied to interpret feature contributions. Results indicate that the XGBoost model achieves the highest predictive performance (R2=0.73, and the average absolute percentage error is 14.0%). The most influential factors affecting alloy hardness are mixing enthalpy (), Nb content, and atomic size mismatch (). Effective hardness control relies on the synergistic regulation of thermodynamic stability, electronic structure, and geometric dimensions, where inter-feature compensation plays a critical role in optimizing overall performance.