Unveiling Price Drivers in Volatile Electricity Markets: An Explainable AI Framework

Abstract

The integration of renewable energy sources has introduced significant volatility into electricity markets, rendering traditional forecasting models insufficient. While advanced Machine Learning models offer superior accuracy, their black-box nature creates a critical trust gap for market participants who require interpretable insights for risk management. This study addresses this challenge by developing an interpretable and explainable AI framework for forecasting day-ahead market prices in emerging economies. Using Türkiye as a representative case study for markets with high currency volatility, we benchmark seven forecasting algorithms. LightGBM emerges as the best model, demonstrating statistically significant superiority via the Diebold-Mariano test. Notably, we utilize SHapley Additive exPlanations (SHAP) to validate the model against economic theory, thereby confirming its ability to learn the non-linear merit order effect independently. By transforming opaque predictions into explainable economic signals, this framework bridges the gap between predictive accuracy and stakeholder trust, offering a robust tool for decision-making in high-stakes energy trading environments.