Enhancing Lithium-ion Battery Performance Through Crystal System Classification: Insights From Structural and Physical Feature Analysis

The physical and chemical characteristics of Li-ion batteries’ (LIBs’) performance are greatly influenced by the crystal structure system. Accordingly, identifying and classifying the crystal structure of LIBs is critical for improving their performance and safety. This study classifies LIBs into three main classes of crystal systems, monoclinic, orthorhombic, and triclinic, utilizing machine learning techniques. The performance of different models, including ensemble and non-ensemble models, was checked on this challenging classification task via key evaluation indicators. Among the different models, the standard Random Forest (RF) model provided a very strong performance; after optimization, this model was further improved to outperform all the other models with the best accuracy, precision, and generalization for both the training and test datasets. Also, the important axis of this work was the role of features in driving the classification performance. Enriched by the intrinsic and derived features, representative of the structural and physical properties of battery materials, models gained an enhanced capability to understand and distinguish crystal systems. All these features became critical for the improvement of model accuracy and interpretability. Sensitivity analysis and SHAP evaluation revealed the fact that Band Gap, Formation Energy, Volume, Density, and Volume to site features have high importance to subtle differences among the three crystal classes. These findings provide leverage in advancing the research into batteries and set a basis for future applications within the classification tasks of material science.