Explainable AI Framework Using XGBoost With SHAP and LIME for Multi-Scale Household Energy Forecasting

Forecasting energy consumption is critical for maximizing resource utilization, lowering operational costs, and increasing sustainability. Traditional machine learning models often function as black boxes, which complicates the understanding of decision-making processes. This research utilizes Explainable Artificial Intelligence (XAI) methodologies to improve the interpretability of Extreme Gradient Boosting (XGBoost) to effectively predict energy consumption patterns. Shapley additive explanations (SHAP) is applied to reveal the significance of both global and local features, thereby clarifying how various factors of energy consumption influence the forecasts. Furthermore, Local Interpretable Model-Agnostic Explanations (LIME) contribute to model interpretability by providing a detailed examination of individual predictions. The proposed model identifies key determinants affecting energy usage through Permutation Feature Importance (PFI), while Partial Dependence Plots (PDPs) illustrate the relationships between input variables and the model’s predictions. Additionally, Holt-Winters Exponential Smoothing is incorporated as a time-series decomposition method to investigate both short and long-term trends to predict energy consumption across various time intervals: hourly, daily, weekly, monthly, quarterly, as well as at the appliance level. This approach achieves a high R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> value of 0.9999 and low RMSE values of 0.00344. Comparisons with baseline models demonstrate the superior performance of the proposed method. By integrating Holt-Winters + XGBoost with XAI (HW+XGB+XAI) techniques, this study enables AI-driven energy management and home energy optimization, ensuring both interpretability and trust.