Decoupling complex cell aging with explainable machine learning: a perspective

Abstract Lithium-ion batteries (LIBs) undergo irreversible and complex aging processes, making it essential to understand the influence of cycling parameters on their State of Health (SOH). Although previous aging studies have systematically varied cycling parameters to assess battery lifetime, the relative importance of these parameters on degradation mechanism with declining SOH is not well characterized. In this perspective, we apply explainable machine learning (ML) to a publicly available dataset of 28 LIBs to quantify the contributions of key cycling parameters on degradation as a function of SOH. Our analysis reveals that in the early stages of life (SOH 0.975–0.90), degradation is primarily governed by temperature, likely due to the abundance of available electrolyte. At later stages (SOH 0.875–0.80), charging current emerges as the dominant driver, reflecting the growth of the solid–electrolyte interphase, electrolyte depletion, and associated effects such as Joule heating. These stage-dependent findings align with prior mechanistic studies and demonstrate how combining small-data ML with interpretability can yield new insights into degradation pathways. Overall, this work provides a qualitative framework for linking cycling parameters with SOH-dependent degradation, offering guidance for strategies to extend the cycle life and safety of LIBs.