A Unified Framework for Interpretable and Uncertainty-Aware Battery State of Health Estimation Using Deep Neural Networks

<p dir="ltr">Accurate estimation of battery state of health (SOH) is essential for the safety and reliability of electric vehicles. While deep neural networks have shown promise in predicting SOH from sensor data, they remain opaque and lack quantifiable confidence estimates. This paper presents a unified framework that integrates a multi-method explainability toolkit comprising SHAP, LIME, integrated gradients, and counterfactual explana?tions. Additionally, we propose an adaptive confidence interval approach that captures both epistemic and aleatoric uncertainty. Evaluations using benchmark models show that our framework improves transparency and safety, achieving a mean prediction interval width of 0.2311, corresponding to 23.11% of the normal?ized SOH range and a prediction interval coverage probability of 0.6288 (62.88% of ground truth values fell within the predicted intervals). These results highlight the potential of combining explainable artificial intelligence and uncertainty quantification to enable trustworthy battery diagnostics.</p>