Phase prediction in high-entropy alloys with multi-label artificial neural network

The prediction of phase composition in metallic alloys is one of the main challenges in modern material science. The most effective and promising methods to solve this problem are currently based on machine learning (ML) algorithms. The most urgent issues in developing such methods are the choice of input features (descriptors) and the search for the most effective ML models. Here we address these issues for the problem of phase composition prediction in high-entropy alloys (HEAs). We combine two ideas recently proposed in this field: the use of genetic algorithms to search for optimal sets of descriptors and a multi-label classification scheme. By using this approach, we achieve the value of balanced accuracy of more than 91% in the prediction of selected phases.

Automated summary

The prediction of phase composition in metallic alloys is a challenge in material science, which can be effectively addressed using machine learning algorithms. This study focuses on the prediction of phase composition in high-entropy alloys and achieves an accuracy of over 91% by combining genetic algorithms and a multi-label classification scheme.