Accurate State of Charge (SOC) estimation is a critical requirement for reliable and efficient Battery ManagementSystems (BMSs) in lithium-ion battery-powered electric vehicles. However, SOC estimation remains challenging becauseofthehighly nonlinear and temperature-dependent behavior of lithium-ion batteries under dynamic operating conditions. Thispaperpresents a comparative machine learning framework for SOC estimation using Ensemble LSBoost Trees and LongShort-TermMemory (LSTM) networks under multi-temperature operating environments. The proposed framework utilizes the TUBerlinlithium-ion battery dataset containing dynamic drive-cycle profiles at 5°C, 15°C, 25°C, 35°C, and 45°C. Acomprehensivepreprocessing and feature engineering pipeline consisting of missing-value handling, outlier correction, Savitzky–Golayfiltering,derivative feature extraction, moving average current computation, and z-score normalization is implementedtoimproveestimation robustness and prediction accuracy. To evaluate thermal generalization capability, the models are trainedusingbatterydata from 5°C–35°C and tested exclusively on unseen 45°C operating conditions. Experimental results demonstrate that theLSTMmodel outperforms Ensemble LSBoost Trees because of its superior temporal learning capability and sequential dynamicmodeling.The proposed LSTM network achieves an RMSE of 0.0363, MAE of 0.0296, and an R2 score of 0.9823, whereas theEnsembleTrees model achieves an RMSE of 0.0486 and an R2 score of 0.9686. Keywords: State of Charge (SOC), Lithium-IonBattery,Battery Management System (BMS), Long Short-Term Memory (LSTM), Ensemble Trees, Machine Learning, DeepLearning,Electric Vehicles