Mechanical Property Prediction of High-Entropy Alloys Using Machine Learning Methodology

High-entropy alloys (HEAs) have been gaining increased attention due to their remarkable properties including excellent mechanical performance at high temperatures, exceptional ductility, high fracture toughness at cryogenic temperatures, high conductivity, and excellent catalytic and magnetic properties. However, the availability of mechanical property data is very limited and sparse in comparison with conventional alloys such as steels and others. In this particular investigation, yield strength prediction was carried out using a modest dataset, consisting of 697 instances of HEAs, sourced from experimental literature. Three supervised machine learning regression models: decision tree regressor (DTR), random forest regressor (RFR), and extra-tree regressor (ETR) were employed in MAterials Simulation Toolkit for Machine Learning (MAST-ML) framework. The methodology of MAST-ML was thoroughly examined along with its constraints. ETR model was observed to perform the best among others, with R2_score (coefficient of determination), mean absolute error (MAE), and root mean squared error (RMSE) of 0.924, 0.09, and 0.148, respectively, for test dataset. Extensive testing of new/unseen compositions that were not the part of either training or test set was conducted to ensure the model’s generalizability. A notable consensus between the predicted and actual yield strength values was observed.