FORECASTING VN-INDEX TIME SERIES BASED ON CEEMDAN DECOMPOSITION AND DEEP LEARNING

Abstract

In this note we propose a model of prediction for VN-Index time series by combining three deep learning techniques including Complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), Temporal convolutional networks (TCN) and Gated recurrent unit neural networks (GRUNN) together. The aim of our research is to reduce the number of components in CEEMDAN decomposition so that it decreases time consumption for training models. Instead of dividing the components in CEEMDAN decomposition into many parts, our proposed model grouping these components into two main groups, one with high permutation entropy and other with low permutation entropy. Empirical results indicate the better performance of the model for predicting a financial time series VN-Index than those of other models.