Machine learning-enabled framework for the prediction of mechanical properties in new high entropy alloys

Prediction of properties of new compositions will accelerate the material design and development. The current study uses a machine learning framework to predict the microhardness of high entropy alloys. Several feature selection algorithms are used to identify the essential material descriptors. The stability selection algorithm gives optimum material descriptors for the current dataset for the microhardness prediction. Eight different machine learning algorithms are trained and tested for microhardness prediction. The accuracy of prediction improved by reducing the higher-dimensional data to lower dimensions using principal component analysis. The current study shows the testing R-2 score of more than 0.89 for XGBoost, Random forest, and Bagging regressor algorithms. Experimental data confirms the applicability of various trained algorithms for property prediction, and for the current study, ANN shows better performance for the new experimental data. (C) 2022 Elsevier B.V. All rights reserved.

Automated summary

This study predicts the microhardness of high entropy alloys using a machine learning framework and demonstrates improved prediction accuracy through feature selection algorithms and dimensionality reduction techniques. Experimental data confirms the applicability of trained algorithms for property prediction.