A hybrid of EMD-SVM based on extreme learning machine for crude oil price forecasting

Abstract

Forecasting crude oil spot prices (COSP) are important as it affects other key sectors of the economy including the stock market. This makes it crucial to develop reliable models that would assist adequately in forecasting the fluctuation of international crude oil price. This is aimed at facilitating the parties involved in taking appropriate action to avoid associated risk. In a current study, a new hybrid method based on empirical mode decomposition (EMD), support vector machine (SVM) and extreme learning machine (ELM) is presented. The crude oil price is adaptively decomposed into a series of smooth intrinsic mode function (IMF) with different scales via EMD. Each extracted IMF was forecasted with different SVM, the final results were obtained by adding together these forecasted results of each IMF. A hybrid method based on an extreme learning machine with adaptive metrics of input is proposed for improving the forecast accuracy of the prediction of all combined IMF. The EMD-SVM-ADD model applies the SVM to predict IMF extracted by EMD and integrates the predicted results, using a simple averaging method. To develop the model, we start by decomposing of the WTI by the EMD extraction process after this for each of the IMF obtained a unique EMD- SVM model is developed which we call IMF-SVM models. The desired EMD-SVM- ADD model is achieved by summing up the predictions from the IMF-SVM. This summing up is where the model derived its name from EMD-SVM because it is simply the additions of the predictions from the smaller IMF-SVM models. To evaluate the efficiency of the model, the study adapts the West Texas International (WTI) crude oil spot price. The results revealed that the new proposed model (EMD-SVM-ELM) performed better when compared with single SVM and EMD-SVM-ADD models judging by their RMSE and MAE. We concluded, based on the results obtained especially with the MAPE is less than 5% that the model is equally suitable for crude oil spot price forecasting.