Machine learning-driven design of low-density Ta-Nb-W-V-Zr-Ti-Mo refractory high-entropy alloys for high-temperature applications

High-entropy alloys (HEAs) are gaining significant attention due to their unique microstructures and outstanding properties. However, traditional design approaches are time-intensive and labor-intensive process, making machine learning (ML) a promising tool for accelerating discovery. In this work, we explored the prediction of density for lightweight refractory high-entropy alloys (LRHEAs), incorporating alloying elements and liquidous and solidus temperature into the analysis. To evaluate the machine learning models, we used numerous performance matrices, together with the coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE). After selecting the optimal model, we successfully predicted the density of new alloys. The XGB model proved to be the most effective, yielding impressive performance metrics (R² = 0.995, MAE = 0.6 %, RMSE = 0.6 %).