A Web-Integrated Machine Learning Approach for SOC Prediction Based on Electrical Impedance and Voltage Measurements

OpenAlex English

Accurate estimation of the State of Charge (SOC) is essential for the efficient and safe operation of lithium-ion battery systems, particularly in electric vehicles and renewable energy applications. This study proposes a machine learning (ML)-based framework for SOC classification using electrical measurements such as impedance components and voltage. Three classification algorithms—Random Forest, Histogrambased Gradient Boosting, and Logistic Regression— were developed and evaluated. The dataset was augmented using Gaussian noise and normalized to mitigate overfîtting and enhance generalization. The Random Forest model achieved the highest performance, with a test accuracy of 96.2% and an F1 score of 0.96. The models were deployed using a user-friendly web interface integrated into the Hugging Face Spaces platform via Streamlit and Docker. Explainable AI (XAI) techniques, particularly permutation-based feature importance, were applied to improve model interpretability. The results indicate that the imaginary part of impedance and voltage are the most influential features in SOC classification. This work demonstrates a robust and interpretable SOC estimation system that is suitable for real-world integration and online accessibility.