Explainable AI models track EV battery health over 6G networks

Researchers have proposed a framework that combines machine learning and deep learning models with explainable artificial intelligence to estimate the state of health of electric vehicle batteries, running the computation across an IoT–Fog–Cloud architecture operating over IoT-based 6G networks. The work, published in Neural Computing and Applications, aims to deliver SOH estimates that are not only accurate but also transparent and trustworthy, addressing a critical factor in ensuring the reliability, safety, and longevity of EV batteries with direct implications for vehicle performance and energy efficiency.

The team systematically evaluated multiple learning models, including deep neural networks (DNN), convolutional neural networks (CNN), long short-term memory networks (LSTM), recurrent neural networks (RNN), and hybrid architectures such as LSTM–CNN, DNN–CNN, and LSTM–KNN.

The experimental results showed that recurrent-based architectures, particularly the LSTM–RNN model, consistently outperformed the other approaches in terms of accuracy and robustness across varying training ratios. The LSTM–RNN model achieved RMSE values as low as 0.0008, the lowest error reported in the study.

Not every hybrid combination fared as well. The study found that hybrid models such as LSTM–KNN exhibited significantly higher errors and reduced reliability, especially under data-limited conditions when only a fraction of the available cycling data was used for training. This contrast suggests that simply combining learning paradigms does not guarantee better performance, and that the choice of architecture matters considerably when training data is scarce.

A central contribution of the work is the incorporation of explainable artificial intelligence techniques, namely LIME and SHAP, into the prediction pipeline. These interpretability methods reveal which input features contribute most to each model’s predictions, allowing researchers and engineers to understand the basis of an SOH estimate rather than treating the model as a black box. Importantly, the authors report that adding these explanation layers did not degrade predictive performance, so transparency was gained without an accuracy penalty.

The proposed IoT–Fog–Cloud architecture is designed to support efficient real-time battery monitoring, data transmission, and predictive analytics. By distributing computation across fog and cloud layers connected through 6G networks, the framework is intended to offer a scalable, low-latency solution for SOH prediction in IoT-enabled EV battery management systems, supporting real-time monitoring and informed decision-making about battery condition.

Accurate SOH estimation matters because it underpins judgments about when batteries remain safe and dependable, how long they will last, and how efficiently a vehicle uses energy. Data-driven SOH prediction has been enabled by machine learning and deep learning methods that effectively capture complex degradation patterns from large-scale battery datasets.

The authors position the framework as a step toward intelligent, explainable, and reliable battery management solutions for next-generation electric vehicles, combining the predictive strength of recurrent deep learning models with the accountability that interpretability techniques provide. As with any study based on benchmark laboratory datasets, performance on the NASA PCoE cells indicates how the models compare under those test conditions, and the reported results demonstrate the approach’s accuracy and robustness within that evaluation scope.

Subject of Research: Explainable machine learning and deep learning models for electric vehicle battery state-of-health estimation in IoT-based 6G networks

Article Title: Explainable AI-driven machine and deep learning models for state-of-health estimation of EV batteries in IoT-based 6G networks

Article References: Sakr, H. A., Fouda, M., Ashour, A. F., Eladl, A. A., & El-Afifi, M. I. (2026). Explainable AI-driven machine and deep learning models for state-of-health estimation of EV batteries in IoT-based 6G networks. Neural Computing and Applications, 38(20), Article 823. https://doi.org/10.1007/s00521-026-12491-4

Image Credits: AI Generated

DOI: 10.1007/s00521-026-12491-4

Keywords: electric vehicles, battery state of health, explainable AI, LSTM, deep learning, LIME, SHAP, 6G networks, IoT, NASA PCoE dataset, battery management, machine learning