Hierarchical machine learning based structure–property correlations for as–cast complex concentrated alloys

George Stephen Thoppil, Jian Feng Nie, Alankar Alankar

Research output: Contribution to journalArticleResearchpeer-review

1 Citation (Scopus)


Efficient design of complex concentrated alloys (CCAs) requires knowledge of constituent elements and their fundamental properties, phase selection, phase-fractions and microstructure and their correlations with properties. We present a novel framework incorporating a feature selection scheme to extricate the factors deciding phase formation and mechanical properties — both compressive and tensile for as-cast compositions. These features have been harnessed using classifier and regressor models to predict the phases and properties of compositions for which they are as yet undocumented. The stacking fault energies and density regressor models augment the above feature set. The phases are predicted using a novel multi-output phase classifier model that could predict up to 4 phases simultaneously with accuracy in the range of (95–99)%. The hardness, yield strength, peak strength and fracture strain regressor models also perform well with testing metrics in the range (0.91–0.96)R2. The regressors are subsequently operated upon an artificially generated composition space of 5-element CCAs comprising a select few elements, for which the properties in the as-cast condition are predicted.

Original languageEnglish
Article number111855
Number of pages14
JournalComputational Materials Science
Publication statusPublished - 5 Jan 2023


  • Complex concentrated alloys
  • High entropy alloys
  • Materials informatics
  • Structure–property relations

Cite this