Machine Learning‐Guided Design of Next‐Generation Battery Materials for Sustainable Energy Storage Systems

ABSTRACT

This study presents a machine learning‐driven framework for predicting material stability and sustainability to accelerate the discovery of environmentally friendly battery materials. Using data retrieved from the Materials Project, key material descriptors were extracted and used to train predictive models, including random forest Regression and XGBoost. The random forest model demonstrated superior performance in predicting formation energy, achieving a mean absolute error (MAE, E _MA ) of 0.2643 and an R ² score of 0.5857, compared to XGBoost ( E _MA  = 0.2669, R ²  = 0.1946). For sustainability index prediction, random forest also outperformed with E _MA  = 0.0810 and R ²  = 0.5012. Further validation using k ‐fold cross‐validation confirmed strong model reliability, with random forest achieving a E _MA of 0.0203 ± 0.0099 and mean R ² of 0.9452 ± 0.0756, surpassing XGBoost ( E _MA  = 0.0246 ± 0.0070, R ²  = 0.8700 ± 0.0777). Feature importance analysis revealed that only a few key descriptors significantly influence predictions, whereas correlation analysis showed that lower formation energy is associated with higher sustainability. The trained models were successfully applied to predict new candidate materials, including Li ₂ O, which exhibited a high sustainability index of 0.9514. Overall, this work demonstrates that machine learning can effectively guide the identification of stable and sustainable materials, contributing to the development of next‐generation energy storage technologies.