Applicability of machine learning approaches for predicting the phase formation in complex concentrated alloys

This work deals with the machine learning techniques used to build predictive models to determine the phases in complex concentrated alloys (CCAs). Two different approaches were employed to determine the presence of phases. The one that relies on thermodynamic parameters was insight invoking majorly, and the other helped to predict phases for the specified alloy composition. Predictions were made using ensemble models and neural networks. Five machine learning (ML) algorithms were applied that are (1) K-nearest neighbors (KNN), (2) Support vector machines (SVM), (3) Gradient Boosting, (4) Ada-Boost Classifier, and (5) Histogram gradient Boosting Classifier. These algorithms were utilized for the prediction of Solid Solution (SS) and Intermetallic (IM) phases in CCAs. Among all the models, the Histogram Gradient Boosting Classifier model and Gradient Boosting model demonstrated the highest accuracies of 85.0 % and 83.7 %, respectively, for phase prediction with thermodynamic parameters. For phase prediction considering the alloy composition, the Histogram Gradient Boosting Classifier model obtained an accuracy of 84.8 %, while the Gradient Boosting model resulted an accuracy of 82.3 %. It was deduced that Histogram Gradient Boosting Classifier is the most significant model with respect to phase predictions in CCAs. • Thermodynamic parameters and composotion based approaches for predicting phase fromation is CCAs. • Esemble models and neaural networks employing five ML algorithms were applied. • Solid Solution BCC, Solid Solution FCC, Solid Solution (BCC+FCC) and Intermetallic (IM) phases were considered. • Histogram Gradient Boosting Classifier is the most significant model with respect to phase predictions in CCAs.