Accurate assessment of battery degradation remains crucial for the reliable operation of lithium-ion battery systems. This study presents a novel approach using Gaussian mixture models (GMM) as a mathematical framework for decomposing incremental capacity curves and extracting features for state of health estimation. The GMM decomposition transforms complex IC curves into a compact parametric representation, with each Gaussian component characterized by position, width, amplitude, and area parameters. These parameters systematically evolve with battery aging, providing rich information for health assessment. We evaluate three machine learning approaches—Gaussian process regression, random forests, and support vector regression—to establish the relationship between GMM parameters and state of health. Using the Oxford Battery Degradation Dataset, Gaussian process regression achieves the best performance with RMSE of 0.64, followed by random forests (1.40) and support vector regression (1.86). The method demonstrates robust performance across different cells with computational efficiency suitable for real-time applications (~0.03s per curve). The systematic feature extraction framework offers a valuable tool for battery diagnostics, balancing accuracy with practical implementation requirements.

Lithium-ion battery,State of health,Gaussian mixture model,Incremental capacity analysis,Battery degradation