A Machine Learning Framework for Predicting Lithium-Ion Battery State of Health Based on Operational Degradation Indicators

Predicting the State of Health (SoH) of lithiumion batteries (LIBs) is essential for maintaining their performance, safety, and longevity-especially as their use expands across energy storage systems. In this study, we introduce a machine learning-based model that leverages real operational data from NASA’s battery dataset. Our framework combines signal filtering, feature extraction, and ensemble regression to model the complex degradation behavior of LIBs. Among the models tested, Random Forest (RF) offered the highest accuracy across MSE, RMSE, MAE, and MAPE. Importantly, cumulative indicators such as Capacity Loss and Ah-Throughput emerged as the most predictive features, underscoring the significance of long-term usage metrics. These results highlight the potential of ensemble learning methods and informed feature selection in developing accurate and practical tools for battery health estimation.