An XGBoost-SHAP Model for Energy Demand Prediction With Boruta–Lasso Feature Selection

Energy demand prediction is essential in ensuring national energy security, promoting high-quality economic development, advancing sustainable development, optimizing the energy structure, and achieving dual carbon goals. In recent years, machine learning (ML) algorithms have been extensively used in energy demand prediction but with poor interpretability. This study proposes an interpretable ML framework for energy demand prediction based on the Boruta-Lasso two-stage feature selection model, extreme gradient boosting (XGBoost) regression model, grid search optimization algorithm, and Shapley additive explanations (SHAP) algorithm. Taking China as an example, the results show that the XGBoost-based model achieves R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>, MAE, and RMSE of 0.9975, 0.0108, and 0.0157, respectively, with higher prediction accuracy compared to other models. In addition, the analysis based on the SHAP algorithm shows that gross national income (GNI), gross domestic product (GDP), total population at the year-end, per capita consumption expenditure, and the proportion of added value of the secondary industry are the main drivers of energy demand. The framework established in this study helps to screen influencing factors, predict energy demand, and simultaneously explain the relationship between these factors and energy demand.