High-entropy oxides are a novel class of materials with promising applications in energy conversion and storage; however, their rational design remains challenging due to the immense compositional space. Here, we propose a machine-learning-based methodology to design stable, single-phase HEOs. We trained predictive models to identify candidate fluorite-structured compositions. The ensemble achieved reasonable performance in a six-class classification task, as evaluated using nested stratified cross-validation and external validation (weighted-average F1-scores of 77% and 70%, respectively). We further applied SHAP analysis to assess the physical relevance of the predictors. Experimentally, we synthesized Ce 0.2 La 0.2 Nd 0.2 Mg 0.2 Al 0.2 O 2 − δ . X-ray diffraction confirmed this prediction, and transmission electron microscopy (TEM) combined with selected-area electron diffraction (SAED) validated the phase assignment. Overall, these results demonstrate that machine learning is a powerful approach to navigate the complex HEO compositional landscape and accelerate the discovery of materials with targeted properties. • Novel machine learning method to design new high-entropy oxides. • Synthesis of a new high entropy oxide, Ce 0.2 La 0.2 Nd 0.2 Mg 0.2 Al 0.2 O 2 − δ . • Electron affinity and electronegativity (across multiple scales) strongly influence phase formation.
Santos et al. (2026) studied this question.