Design of novel full Heusler alloys Mn₂YAl (Y[dbnd]Fe, Sc, Ni, co): A combined machine learning and DFT study of magnetic and electronic properties

This study presents a novel approach for predicting the magnetic and electronic properties of Full Heusler alloys (FHAs) by employing eXtreme Gradient Boosting (XGBoost), a machine learning technique, alongside Density Functional Theory (DFT) calculations. The model employs feature engineering, k-fold cross-validation, and Bayesian optimization (BO) to precisely predict the material properties derived from an extensive library of Heusler alloy compositions. we employed rigorous assessment measures, resulting in a high performance on the unseen test set with R² = 0.90, MSE = 0.13, and MAE = 0.21 for the magnetic and R² = 0.87, MSE = 0.14, and MAE = 0.26 for the electronic properties. Compared to DFT, the XGBoost model produces very precise predictions with negligible error from −12.4 to 9.6 %, thus increasing the rate at which new materials are discovered. This work demonstrates the application of machine learning (ML) in materials science and facilitates further exploration of HAs, which are characterized by their magnetic and spintronic properties.