A Comparative Study of Machine Learning Methods for Lithium-ion Battery Remaining Useful Life Prediction

Abstract

Accurate prediction of remaining useful life for lithium-ion batteries is essential for ensuring the reliability and safety of electric vehicles and energy storage systems. This study presents a comprehensive comparative analysis of six machine learning methods for battery RUL prediction, including Linear Regression, Ridge Regression, Lasso Regression, Support Vector Regression, K-Nearest Neighbors, and Random Forest. Using the NASA battery degradation dataset, we develop a feature engineering framework that extracts six capacity-derived features while carefully avoiding information leakage by excluding cycle number from the feature set. A Leave-One-Battery-Out cross-validation strategy is employed to provide realistic generalization estimates for time-series degradation data. Experimental results demonstrate that Lasso Regression achieves the best predictive performance with a root mean square error of 10.827 cycles and a coefficient of determination of 0.8183 on the test battery. The findings reveal that linear regularized methods outperform more complex nonlinear approaches for this prediction task, suggesting that the relationship between capacity-derived features and RUL is predominantly linear. This work provides practical guidance for selecting appropriate machine learning algorithms for battery health management applications.