<![CDATA[The research will enhance the forecasting of the lithium-ion battery degradation to facilitate more secure and sustainable energy storage in the electric vehicle. An analytical framework that is hybrid in nature, incorporating both statistical analysis and artificial neural network (ANN) modeling, was developed and verified using a long time dataset of INR21700- M50T cells being cycled in realistic urban driving profiles according to the Urban Dynamometer Driving Schedule (UDDS). The indicators of key degradation were first described using statistical analysis where it was found that there were strong negative relationships between capacity retention and capacity C-rate (Pearson r = -0.83) and internal resistance (r = -0.71). Based on these findings, a feedforward neural network, whose features were selected using ReliefF algorithm, was built which was used to model nonlinear aging behavior at lower input dimensionality. ANN inputs were chosen as the 2 most powerful features low-frequency impedance at 0.01 Hz and internal resistance. The resulting model had a high predictive performance of a root mean squared error (RMSE) less than 1.2% and a coefficient of determination (R2) greater than 0.97 on the original data. These results underscore the fact that combining data-based feature relevance analysis with machine learning is useful in improving the accuracy of prediction as well as the interpretability of the model. The obtained results demonstrate that combining statistically supported feature relevance analysis with reduced-input ANN modeling can improve both predictive capability and model interpretability for battery degradation estimation. The proposed hybrid framework provides a computationally efficient approach for lithium-ion battery state-of-health prediction under the investigated dataset conditions and may support future development of simplified battery management system strategies.]]>
