Explainable Machine Learning based approach for the design of new refractory high entropy alloys

The design and development of Refractory High Entropy Alloys can be expedited using Machine Learning (ML) approach. The current study uses a RHEA database to train the different Machine Learning models, namely, Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), K-nearest neighbor (KNN) and Artificial Neural Network (ANN). The trained models show an average testing accuracy of more than 80 %. Local interpretable model-agnostic explanations (LIME) and Shapley additive explanations (SHAP) are used to understand the impact of design parameters on ML phase prediction. New RHEAs are predicted by trained ML models and verified by experiments. The significant design parameters which affect the ML prediction of new alloys were established using LIME.