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

Forecasting household energy consumption is important for understanding patterns and improving energy management. This project presents a hybrid forecasting approach combining Holt-Winters Exponential Smoothing, GRU/LSTM networks, XGBoost, and Random Forest to predict energy usage at multiple scales. The Holt-Winters method decomposes the time series into trend, seasonal, and residual components, allowing the model to capture both short-term fluctuations and long-term trends. GRU/LSTM networks are used to model temporal dependencies and generate features that are fed into XGBoost and Random Forest regressors for final prediction. To improve interpretability, LIME is applied to explain individual predictions, highlighting the contribution of each feature. The proposed hybrid model demonstrates high accuracy, low error metrics, and reliable performance across different time intervals. By integrating time-series decomposition, deep learning, and ensemble methods with explainable AI, this approach provides accurate and interpretable energy forecasts. The results show that combining multiple models enhances prediction capability while ensuring feature-level transparency.

Keywords: Household Energy Forecasting, Explainable AI, XGBoost, Random Forest, GRU, LSTM, Holt-Winters Exponential Smoothing, LIME, Time-Series Analysis, Multi-Scale Prediction.