Accurate energy power prediction has become increasingly critical for modern power systems, enabling efficient grid operation, optimal resource allocation, and substantial cost savings. This paper presents a comprehensive comparative review of energy power prediction methodologies, focusing on Light Gradient Boosting Machine (LGBM) and a hybrid approach combining Extreme Gradient Boosting (XGBoost) with LGBM. Ensemble learning methods have demonstrated significant potential in improving prediction accuracy through bias-variance trade-off techniques. However, there remains considerable room for improvement in forecasting performance. This review examines the theoretical foundations, methodological approaches, and empirical performance of various gradient boosting techniques for electrical load forecasting. The study analyzes research conducted on the Pennsylvania-New Jersey-Maryland interconnection power grid dataset, evaluating models based on mean absolute percentage error, root mean squared percentage error, and mean absolute error. The findings reveal that hybrid approaches consistently outperform single models, with the XGBoost-LGBM hybrid demonstrating superior accuracy by reducing prediction errors by more than one percent compared to individual models. This paper synthesizes current literature, identifies research gaps, and proposes directions for future investigation in energy power prediction using advanced ensemble methods.