Machine Learning Helps Data Mining to Build Descriptor Databases for Lithium-Ion Batteries

Abstract

The rapid development of computer science has made machine learning (ML)-driven design a research hotspot in high-performance lithium-ion batteries. Descriptors play a critical role in ML processes, as accurate descriptors significantly improve prediction accuracy (achieving over 92% validation accuracy in density functional theory [DFT]-calibrated models). Although open-source databases offer rich material data, their operational complexity hinders effective utilization. This review highlights ML algorithms that streamline data extraction from these repositories, slashing experimental iterations by 75%–80%. Further analyze future challenges in descriptor acquisition for lithium-ion batteries. This review is to provide insights into dataset construction and ML-compatible descriptor generation, accelerating electrode material discovery from conventional 5–7 years to < 18 months in recent cases.