Machine Learning-Based Computational Design Methods for High-Entropy Alloys

High-entropy alloys (HEAs) have attracted much attention due to their excellent properties and wide range of applications, but their large compositional space and complex property relationships pose challenges to traditional design methods. Machine learning (ML) has become a powerful tool for accelerating the HEA design due to its powerful data processing and prediction capabilities. This review first emphasizes the importance of constructing high-quality datasets for training reliable ML models and analyzes the impact of data quality on model performance. The potential benefits of text-mining techniques in discovering novel HEA candidate materials from large amounts of data were concerned. Based on the data-preprocessing process, the constructions of new descriptors are described in detail, and the uses of domain knowledge to assist in predicting complex HEA performance and to improve the interpretability of ML models are elaborated. The principles, strengths, and weaknesses of various ML models (e.g., support vector machines, decision trees, and deep learning) and their applications in phase selections and mechanical performance are illustrated in detail, as well as the utility of active learning, transfer learning, and inverse-design techniques in guiding the design of experiments. In addition, this review summarizes the cases of ML used in predicting HEA corrosion and oxidation resistance with complex mechanisms. Potential research prospects, such as the extension of reliable data sources, the development of advanced models, and the interpretability of models, are also discussed. This review aims to provide a comprehensive ML guide for HEA researchers and to facilitate the application of ML in further accelerating HEA development. © The Author(s) 2025.