Machine learning-enabled identification of new medium to high entropy alloys with solid solution phases

Identification of compositions that form solid solution phases from a large compositional space is challenging. The current study focuses on applying multiple machine learning classification algorithms for suitably predicting new medium to high entropy alloys with solid solution phases. The current data set used for training and testing consisting of 664 labeled data with 267 BCC alloys, 199 FCC alloys, and 198 (FCC + BCC) alloys to avoid biased predicting. The analyzed data shows a strong correlation between the empirical design parameters. The correlation coefficient values changed while moving the alloy system from medium to high entropy domain. The parameters VEC and Tm show high importance in prediction compared to other parameters. The importance of design parameters is analyzed, and the accuracy is quantified. The experimental results validate the prediction of a shift from BCC + FCC to FCC phases while increasing Ni content in the CoCuFeNix system.