Application of Machine Learning Techniques in Natural Gas Price Modeling. Analyses, Comparisons, and Predictions for Romania

The current global energy crisis is an important topic, which emphasizes the need to study the natural gas market, with appropriate modeling methods, for a proper substantiation of the public policies. The specialized literature is generous in terms of the analyses carried out on the electricity market, but the natural gas market is not a subject fully exploited by researchers; therefore, this article represents an important contribution to knowledge in the field. This article analyses the natural gas market in Romania between November 2016 and September 2022, using data collected daily, representing the weighted average daily price of natural gas. The research is carried out with the help of advanced machine learning methods, namely, a series of basic algorithms (models), but also three categories of ensemble learning methods (bagging, boosting, and stacking). It was found that the price of natural gas in Romania can be estimated with high accuracy, using decision tree (DT) algorithms or with the help of artificial neural networks (ANNs). However, ensemble learning-based modeling proves to be the best estimation method, characterized by reduced prediction errors compared to basic models.