Small features, big impact: Influence of overlooked microstructural features on high-entropy alloy predictions

High-entropy alloys (HEAs) exhibit a range of superior properties, making them attractive candidates for advanced engineering applications. However, their design and optimization remain challenging due to the complex interactions among multiple alloying elements. Machine learning (ML) has emerged as a powerful approach for accelerating the discovery of HEAs with tailored properties. In this study, ML models were employed to predict the hardness of AlCrFeNiTi HEAs across a broad compositional space. By incorporating microstructural knowledge into dataset construction, a more representative and physically informed training set was established. The influences of individual alloying elements on phase stability and microstructure evolution were systematically analyzed, providing a solid foundation for accurate predictions. Notably, the omission of minor microstructural features, such as nanoscale precipitates that occur only within limited compositional ranges, was found to markedly degrade prediction accuracy. We demonstrate a simple yet effective way or strategy to identify these critical features by analyzing compositions associated with the largest prediction deviations. Incorporating a small number of representative compositions containing these features significantly improved model performance. Beyond enhanced predictive capability, this approach also yielded mechanistic insights into the strengthening roles of specific elements in HEAs. These findings underscore the importance of integrating microstructural understanding into ML-driven alloy design to enable more reliable and efficient materials development.