Deep Neural Network Model for Improving Price Prediction of Natural Gas

Natural gas accounts for one of the most industriously marketed energy commodities with a meaningful impact on various financial activities around the world. As direction of price for natural gas changes over time, accurate price prediction of natural gas is essential since this prediction is useful in decision making, commodity marketing, and sustainability planning. In this paper, a deep neural network (DNN) model for monthly price prediction of natural gas is proposed. Deep neural networks are becoming the standard tools that offer a lot of values to researchers for solving different problems in the machine learning and data science community due to their ability for increasing model accuracy. The proposed DNN model presented in this paper utilizes the capability of fully connected layers for learning the dynamics in natural gas price data and the efficiency of Rectified Linear Unit (ReLU) function for performing threshold operations on each input element. A wide range of monthly data covering 281 months were used to develop and test the predictive capability of the proposed DNN model. In comparison to five recently reported mainstream machine learning models, overall results disclose that the proposed DNN model demonstrates superior performance over the mainstream machine learning models with mean squared error (MSE), root mean squared error (RMSE), and coefficient of determination $({\mathrm {R}}^{2})$ of 0.0595, 0.2440 and 0.9937, respectively.