Electricity price forecasting with ensemble meta-models and SHAP explainers: a PCA-driven approach

Abstract Accurate electricity price forecasting is essential for optimizing market operations, enhancing resource allocation, and ensuring sustainable energy management in volatile and complex markets. This research introduces a comprehensive ensemble meta-modeling framework that integrates machine learning techniques with SHAP (SHapley Additive exPlanations) for enhanced interpretability and PCA (Principal Component Analysis) for effective dimensionality reduction. The methodology capitalizes on the complementary strengths of predictive models such as XGBoost, LSTM, and CNN to address the non-linear and temporal intricacies of electricity price datasets. Two ensemble approaches were implemented: (1) Weighted Averaging, assigning weights inversely proportional to model RMSE, achieving an RMSE of 2.126761, and (2) Meta-Model Ensemble, employing Linear Regression, achieving superior accuracy with an RMSE of 1.939032. SHAP analysis provided actionable insights into model contributions, highlighting XGBoost and LSTM as key components. Furthermore, error trajectory analysis demonstrated the robustness of the ensembles in minimizing cumulative forecasting errors over time. This study contributes to the field by combining advanced machine learning models, ensemble strategies, and explainability frameworks to deliver an interpretable, high-performing electricity price forecasting system. The results inform policy-making and lay the foundation for scalable, data-driven energy market solutions.