Implementing Method of Empirical Mode Decomposition based on Artificial Neural Networks and Genetic Algorithms for Crude Oil Price Forecasting

Abstract

Fluctuations in crude oil prices can affect a country's economic policies. The movement of crude oil prices tends to be nonlinear and non-stationary. One forecasting method that is intended to accommodate these traits is forecasting that integrates empirical mode decomposition (EEMD) ensemble methods based on artificial neural networks and genetic algorithms. In the EEMD method, a white noise signal is added to compensate for the mixture mode that can be formed. Each IMF and residue generated in the decomposition process are used as input to a feedforward neural network (FNN) artificial neural network to obtain forecasting models from each IMF and residue. The genetic algorithm is integrated with the FNN to avoid overfitting, the formation of local optima solutions, and the sensitivity of the selection of FNN parameters. The data in this study uses West Texas Intermediate (WTI) and Brent oil prices. The results of the performance comparison trials for several combination forecasting methods can be concluded that the forecasting results that integrate the EEMD method with JST-GA provide better results compared to the forecasting method that integrates EMD with ANN and EEMD with ANN. The forecasting method developed in this study resulted in forecasting with RMSE / Dstat values of 0.0257 / 61.5936% and 0.0270 / 72.0930% respectively for daily and monthly data from WTI oil types; and the RMSE / Dstat value of 0.0229 / 58.8128% and 0.0300 / 81.5789% respectively for daily and monthly data from the type of Brent oil.