Synergistic prediction of phase and hardness in high-entropy alloys via the integration of machine learning and active learning strategies

Abstract

Due to the extensive compositional space and limited sample size, achieving comprehensive and accurate predictions of the microstructure and properties of high-entropy alloys (HEAs) using machine learning models remains a challenging task. To address this issue, this study proposes a machine learning framework that integrates feature engineering with an active learning strategy, enabling the synergistic classification of phase and hardness grade in HEAs. A dataset was constructed by consolidating experimental data from the literature, and key materials descriptors are identified through a four-step feature selection process. Then a high-precision classification model was developed using the LightGBM model. Furthermore, an uncertainty-based active learning strategy combined with a clustering algorithm was introduced to prioritize samples with the highest information entropy for model improvement, resulting in an increase in the test-set F1_score from 0.81 to 0.85. The results demonstrate that the improved model exhibits stronger consistency with experimental data across different HEA systems, confirming its effectiveness and generalization capability. This study shows that the proposed framework provides an efficient approach for the synergistic prediction of the phase and hardness grade in HEAs, offering new insights and methodologies for the design and optimization of HEA systems.