Phase Prediction in High Entropy Alloys by Various Machine Learning Modules Using Thermodynamic and Configurational Parameters

The purpose of this investigation is to predict the different phases present in various high entropy alloys and subsequently classify their crystal structure by various machine learning algorithms using five thermodynamic, configurational and electronic parameters, which are considered to be essential for the formation of high entropy alloy phases. Proper prediction of phases and crystal structures can eventually trace the properties of high entropy alloy, which is crucial for selecting the suitable elements for designing. The model has been developed by various machine learning (ML) algorithms using an experimental dataset consisting of 322 different HEAs, including 258 solid solution (SS), 31 intermetallic (IM), and 33 amorphous (AM) phases. The ML algorithms include (1) K-nearest neighbours (KNN), (2) support vector machines (SVM), and (3) logistic regression (LR), (4) decision tree (DT), (5) random Forest (RF), and (6) gaussian naive bayes classifier. Among them, both DT and SVM algorithms exhibited the highest accuracy of 93.84% for phase prediction. Crystal structure classification of SS phases was also done using a dataset consisting of 194 different HEAs data, including 76 body centered cubic (BCC), 61 face centered cubic (FCC) and 57 mixed body-centered and face-centered cubic (BCC + FCC) crystal structures and found that the SVM algorithm shows the highest accuracy of 84.32%. The effect of the parameters on determining the accuracy of the model was calculated and tracking the role of individual parameters in phase construction was also attempted.