Enhancing model accuracy in lithium-ion battery crystal system development

One of the most significant contemporary energy storage technologies is lithium-ion batteries, and the performance and stability of such batteries are closely associated with the constituent minerals’ crystal structure. In this work, machine learning is applied for cattregorizing lithium-ion batteries into three major crystal systems: monoclinic, orthorhombic, and triclinic. The performance comparison of several machine learning models, namely Support Vector Classifiers, Nu-support Vector Classification, K-Nearest Neighbors, Adaptive Boosting, Extra Trees, and the optimized Extra Trees model, POA-ET, which is achieved via applying Pelican Optimization Algorithm, are considered in this investigation. Among those, the POA-ET emerges as the best and yields high performances in terms of several key metrics regarding the test and training data sets. The model’s reliability is further validated through further analysis, demonstrating its robustness and generalizability. Spacegroup, Nsites, and Volume respectively are the most important features according to feature importance analysis, significantly impacting the predictions. These results not only demonstrate the predictive capability of the machine learning models but also give insights into the factors underlying lithium-ion battery material performance. The presented work bridges computational modeling with practical applications and thus opens an avenue for creating high-performing lithium-ion batteries in order to aid in the world’s shift to sustainable and clean energy systems.