Machine learning approaches to predicting energy price correlation: From a responsible AI perspective

This study is positioned within Responsible AI practice in energy markets, which exhibit inherent volatility and complexity. We integrate classical and modern machine learning techniques for enhanced energy price correlation forecasting. Principal Component Analysis (PCA) is employed for dimensionality reduction to identify underlying factors driving energy price correlations, leveraging its interpretability as a key analytical advantage. Long Short-Term Memory (LSTM) networks are then introduced for time-series modeling of energy prices and their inter-correlations. Using a controlled simulation experiment, we empirically compare PCA-based and LSTM approaches in predicting energy price co-movements. While PCA provides transparent insights into correlation structure with low computation cost, LSTM achieves higher predictive accuracy (8.7% lower MES, 11.4% lower MAE) by capturing nonlinear temporal dependencies. The analysis highlights a governance-performance trade-off between PCA's interpretability and deep learning's precision, suggesting that model choice should be aligned with institutional capacity, regulatory requirements, and deployment constraints. These findings have significant implications for a technology-driven circular economy transitions, demonstrating how improved predictive modeling can enhance renewable integration and energy efficiency in energy markets. • Integrates classical and modern AI techniques (PCA and LSTM) for dynamic energy price correlation forecasting. • Expands Responsible AI framework using affordance theory, emphasizing explainability, robustness, and accountability. • Demonstrates interpretability–accuracy trade-off through empirical comparison using synthetic energy market data. • Proposes hybrid PCA–LSTM models as Responsible AI-driven solutions for sustainable energy systems and circular economy.