Reducing experimental dependency: Machine-learning-based prediction of Co effects on the mechanical properties of AlCrFeNiCox high-entropy alloys

Abstract

This study applies machine learning (ML) methods, including XGBoost (XGB), random forest (RF), K-nearest neighbors (KNN), support vector regressor (SVR), and linear regression (LR), to predict the mechanical behavior of AlCrFeNiCox eutectic high-entropy alloys (EHEAs) with varying Co content. The objective is to reduce the dependence on experimental data, enabling the study of novel compositions more efficiently. XGB, RF, and KNN emerged as the top-performing models, achieving R2 values of 0.999, with their predictions closely aligned with experimental results. A new stress–strain curve was generated for a composition with a 0.6 molar ratio of Co, where XGB, RF, and KNN achieved R2 values of 0.996, 0.994, and 0.993, respectively. This ML-based approach significantly reduces the need for experimental testing, saving time, cost, and energy while accelerating the development of high-entropy alloys (HEAs).