<![CDATA[Accurate forecasting of household energy consumption plays a crucial role in optimizing energy efficiency, supporting sustainable policy decisions, and improving operational management in smart grid systems. This study enhances conventional XGBoost-based forecasting by integrating cross-validation and residual-based evaluation to ensure model robustness and interpretability. Using a dataset of over 90,000 daily household energy records that include temperature, humidity, and appliance-level usage, a systematic preprocessing pipeline was applied—comprising data cleaning, normalization, temporal feature transformation, and partitioning into training and testing subsets. The proposed model was trained using 10-fold cross-validation to minimize overfitting and validated through residual error analysis to assess stability and bias. Evaluation metrics, including Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the coefficient of determination (R²), demonstrate superior predictive accuracy, achieving MAE = 0.48, RMSE = 0.64, and R² = 0.9864. Visualization of actual versus predicted consumption and symmetric residual distribution further confirm the model’s reliability. The findings highlight that the enhanced XGBoost model not only achieves high precision but also provides a robust foundation for real-time energy monitoring, anomaly detection, and sustainable household energy management. Future work will integrate SHAP-based interpretability and comparative benchmarking with deep learning approaches.]]>
