AI-Based EV Battery Health Prediction with Explainable AI and Database Integration

This research paper presents an AI-based framework for predicting the State of Health (SOH) of electric vehicle (EV) lithium-ion batteries by integrating machine learning, Explainable Artificial Intelligence (XAI), and relational database technology. The proposed framework compares five machine learning regression models: Ridge Regression, Support Vector Regression (SVR), Multilayer Perceptron (MLP), Random Forest, and Gradient Boosting. SHapley Additive exPlanations (SHAP) are used to explain model predictions and identify important battery health indicators, including dV/dQ peak height, internal resistance, and cycle count. The study demonstrates the proposed pipeline using a synthetic, physics-inspired dataset containing 40 battery cells and 28,201 cycle records. Gradient Boosting achieved a Mean Absolute Error (MAE) of 1.26 percentage points, a Root Mean Squared Error (RMSE) of 1.70 percentage points, and an R² score of 0.967 on the held-out synthetic test cells. An SQLite database stores battery information, cycle data, predictions, explanations, and alerts. The framework aims to improve the interpretability, traceability, and management of EV battery health predictions. However, the reported results are based entirely on synthetic data and do not establish real-world battery prediction accuracy. Future work includes validation using public battery datasets, such as NASA and CALCE, and further evaluation under real-world conditions.