Machine Learning Prediction of Single‐Phase Formation in High‐Entropy Boride Ultra‐High Temperature Ceramics

ABSTRACT High‐entropy ultra‐high temperature ceramics (HE‐UHTCs) are promising materials for extreme‐environment applications, but predicting compositions capable of forming stable single‐phase solid solutions remains challenging. In this work, a machine learning framework was developed to predict single‐phase formation in high‐entropy diborides using a dataset of 325 compositions, including 72 single‐phase and 253 multiphase systems. An initial set of 28 physics‐informed descriptors was refined through correlation analysis to 10 descriptors representing thermodynamic, electronic, structural, and compositional effects. Artificial Neural Network (ANN), Support Vector Machine (SVM), and Random Forest Classifier (RFC) models were developed using stratified 10‐fold cross‐validation within the training set and subsequently evaluated on a held‐out 20% test set. All models achieved mean cross‐validation accuracies above 95%, with RFC achieving the highest accuracy (98.46 ± 1.44%), followed by SVM (97.69 ± 2.24%) and ANN (95.33 ± 1.67%). SHAP analysis identified DEED, VECDiff, Allen Electronegativity Difference, and VEC as the most influential descriptors, showing that single‐phase stabilization depends on a combination of thermodynamic favorability, electronic compatibility, and structural compatibility.