Improved Natural Gas Price Prediction Using Random Forest Algorithm

Abstract

The volatile nature of natural gas prices in the energy market has caused concern among various stakeholders who seek to minimize uncertainty by predicting prices in advance. The US Energy Information Administration's maintenance of up to date records on Henry Hub natural gas has made it easier to obtain relevant data for energy market forecasting. Machine learning algorithms have been employed to make accurate price predictions, including Artificial Neural Network (ANN), Support Vector Machine (SVM), gradient boosting machines (GBM), Gaussian process regression (GPR), and least square regression boosting. However, Random Forest (RF) has not yet been used on the Henry Hub dataset. In this study, Henry Hub data was collected from the EIA repository and standardized and validated using 10 fold cross validation techniques. Random Forest was used to predict natural gas spot prices on the standardized dataset, and the results indicated that RF performed better with minimal Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Square Error (MSE) of 0.2756, 0.1814 and 0.0760 than ANN which shows relatively high RMSE, MAE and MSE of 0.607, 0.4328 and 0.685 respectively.