Machine learning-based melting congruency prediction of binary compounds using density functional theory-calculated formation energy

We present the development of a machine-learning (ML) model for predicting the congruency of compound melts by utilizing a combination of density functional theory-calculated formation energies and a database of experimental melting reactions. Among the various ML models tested, the XGBoost model is found to be the most suitable. Therefore, the XGBoost model is trained with a labeled compound database to determine whether a compound melted congruently. Feature importance from the trained model is extracted and compared to identify descriptors crucial for predicting melting behavior. Notably, the change in slope of the DFT convex hull at the composition of the compound (i.e., the “convex hull sharpness”) exhibited significantly higher importance than the other features. The methodology employed in this study has the potential to enhance the design of alloy processing and can be applied to more complex higher-order alloys.