Data-driven discovery of Mn-based full Heusler Alloys: Machine learning and DFT insights into elastic, thermal, and mechanical properties

This study focuses on an innovative approach to design Mn-based Full Heusler Alloys (FHAs) with target elastic, mechanical, and thermal properties for design and validation. The workflow combines Machine learning (ML) prediction, Bayesian optimization (BO), and DFT validation to propose experimentally feasible Mn₂-based FHA candidates. A dataset was created to train ML models, particularly the XGBoost regressor, which exhibited strong predictive accuracy. The model got R² values higher than 0.80 for all properties, with minimal MSE and MAE. Q-Q plot validation, authorizes the normality of residuals, underscoring the statistical strength of our predictive models. Further, we used SHAP values to figure out how important each feature was, which helped us find the most important ones which influenced the properties. Then, (BO) was implemented to design new alloy compositions, which were later validated with DFT simulations. The predictions demonstrated a very close agreement to the simulation results. This shows that combining ML and DFT can speed up the discovery of new materials and make the design of high-performance Mn-based FHAs for a variety technological application.