Explainable deep learning for multi-country energy forecasting and sustainability analysis using climate and socio-economic indicators

Climate change and socio-economic factors have made predicting energy demand and assessing sustainability important problems in today’s energy systems. This work proposes an integrated framework for multi-country energy forecasting and sustainability assessment using deep learning and explainable artificial intelligence (XAI). A multi-country dataset for the years 2020–2024 is used, including climate, socio-economic, environmental and energy-system variables. A number of forecasting models, such as Feedforward Neural Network (FNN), Wide & Deep Multi-Layer Perceptron (MLP), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Deep Autoregressive (DeepAR), are compared. These results show the best performance for the Wide & Deep MLP with a Coefficient of Determination (R²) value of 0.9927, Mean Absolute Error (MAE) value of 0.0158, Root Mean Square Error (RMSE) value of 0.0228, Mean Absolute Percentage Error (MAPE) value of 3.3%, and Willmott’s Index (WI) value of 0.9918. The Wilcoxon signed-rank test is used to assess statistical significance, indicating that the performance improvements over the competing models are significant ( p < 0.05). SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) show that temporal features are the most important predictors, and that climate and socio-economic features offer additional predictive information. Furthermore, a sustainability assessment framework is developed to evaluate transition readiness, decarbonization potential, eco-efficiency, and sustainability performance at the country level. The proposed framework is interpretable and scalable and can be used for energy forecasting and sustainable energy planning.