Interpretable Machine Learning‐Driven Discovery of Sodium‐Ion Battery Cathode Materials With First‐Principles Validation

This work presents a machine learning‐accelerated high‐throughput framework for the discovery of cathode materials for sodium‐ion batteries. A dataset of 4156 sodium‐based battery compounds was constructed from the Materials Project Battery Explorer and represented using 107 structural, chemical, and electronic descriptors. To address incomplete descriptor availability in large chemical spaces, a two‐step learning strategy was employed, where a random forest model was first used to predict missing features prior to property estimation. Multiple machine learning models were benchmarked, with light gradient boosting machine (LGBM) exhibiting the highest predictive accuracy and further optimized using the sparrow search algorithm (SSA) to form an SSA‐LGBM hybrid model. The optimized framework was subsequently applied to screen 12,823 sodium‐containing compounds for average voltage and specific capacity. Initial screening identified 70 candidate materials based on practical performance thresholds, which were further refined using criteria, including thermodynamic stability, electronic conductivity, elemental abundance, cost, and environmental considerations. Ultimately, 15 promising cathode candidates were identified and validated through formation energy and bandgap analysis, as well as comparison with established sodium‐ion battery cathodes. This study demonstrates a scalable and interpretable strategy for accelerating cathode discovery in sodium‐ion battery systems.