Machine learning‐based crystal structure prediction for high‐entropy oxide ceramics

Abstract Predicting the crystal structure is essential to address the reliance on serendipity for facilitating the discovery and design of high‐performance high‐entropy oxides (HEOs). Here, three classic algorithms‐based machine learning models to predict the crystal structure of HEOs are successfully established and analyzed by combining five metrics, and the XGBoost classifier shows excellent accuracy and robustness with ACC and F1 scores up to 0.977 and 0.975, respectively. SHAP summary plot indicates that the anion‐to‐cation radius ratio ( r A / r C ) has the greatest impact on crystal structure, followed by difference in Pauling and Mulliken electronegativities (Δ χ Pauling and Δ χ Mulliken ). It is noteworthy that the r A / r C , Δ χ Pauling, and Δ χ Mulliken lower than 0.35, 0.1, and 0.2, respectively, tend to lead to a fluorite crystal structure, whereas rock‐salt and spinel crystal structures are always formed. This work is expected to facilitate the discovery and design of HEOs with tailorable crystal structures and properties.