Hardness prediction of high entropy alloys with machine learning and material descriptors selection by improved genetic algorithm

With the coming of the age of artificial intelligence and big data, machine learning (ML) has been showing powerful potentials for properties prediction of materials. For achieving satisfying prediction performance, rational feature selection plays a key role along with a suitable ML model itself. In the present work, the traditional genetic algorithm (GA) has been further improved to serve as a feature selection method for the hardness prediction problem of high entropy alloys (HEAs). The concepts of feature importance and gene manipulation were introduced into the improved GA to make it more comprehensible. Comparative analysis demonstrated that the improved GA is superior to the traditional GA in the aspects of accuracy, stability and efficiency obviously. A comparison with other typical feature selection methods was also made. In addition, ML model selection was discussed with the composition feature or the optimal physical feature combination selected by the improved GA. Finally, in order to elevate the prediction ability of ML model, the stacking method as an ensemble learning strategy was proposed in Al-Co-Cr-Cu-Fe-Ni HEAs hardness prediction. It was shown that the prediction errors are successfully lowered. This ML framework could be regarded as a method with general applicability to select suitable ML model and material descriptors, for designing various materials with excellent properties and complex composition.