The Determinants of Electricity Prices through Explainable Machine Learning

Explainable machine learning is used to find the main determinants of electricity prices in Germany. The random forest model is combined with model-agnostic explainers to model and decompose the target time series and is contrasted with the standard linear regression model. The regression random forest model predicts energy prices in Germany with great accuracy. The SHAP values decompose the predictions into contributions from various features, revealing that coal prices and CO₂ ETS (Emissions Trading Scheme) emission prices are the main drivers of energy prices. The model outperforms a linear regression model in terms of accuracy and provides insights into the factors influencing energy prices. The regression random forest model, which is usually considered a black-box model, achieved excellent results in predicting electricity prices in Germany. In contrast, the linear regression model, praised for its interpretability and explainability, produced predictions of lower quality and biased explanations. It is argued that simple models like linear regression may be interpretable and explainable. However, their predictions may be inaccurate, highlighting the importance of balancing prediction quality and explanation accuracy.