Machine Learning-Based Intelligent Prediction of Electric Vehicle Battery Health for Enhanced Per-formance and Lifetime Estimation
Authors-CH.Sravan Kumar, Palle Vasantha
Keyword-Electric Vehicles (EV), Battery State of Health, Battery Management System, Machine Learning, XGBoost, LightGBM, Battery Performance Prediction, Lithium-Ion Battery, Predictive Analyt-ics
The increasing adoption of electric vehicles (EVs) has created a growing demand for reliable battery health monitoring systems that can improve operational efficiency and extend battery service life. One of the most important performance indicators of a lithium-ion battery is its State of Health (SOH), which reflects the remaining capacity and overall condition of the battery throughout its lifecycle. Accurate SOH estimation enables timely maintenance, minimizes unex-pected failures, and improves the reliability of electric transportation systems. This research presents a machine learning framework for predicting EV battery SOH using two ensemble learning algorithms, namely Extreme Gradient Boosting (XGBoost) and Light Gradient Boost-ing Machine (LightGBM). The proposed framework utilizes battery operating parameters in-cluding voltage, current, temperature, and charging history to develop predictive models capable of learning battery degradation patterns. The collected dataset undergoes preprocessing, feature engineering, and train-test partitioning before model training. Performance evaluation is carried out using statistical measures such as R² Score, Adjusted R², Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). Experimental findings demonstrate that both algorithms achieve excellent prediction accuracy, with XGBoost showing slightly superior performance compared to LightGBM. The developed system provides an effi-cient solution for real-time battery health assessment, supporting intelligent maintenance schedul-ing, reducing operational costs, and improving battery lifespan in modern electric vehicles.
Doi-[https://doi.org/10.5281/zenodo.21735381]