A Singular Spectrum Analysis Precoded Deep Learning Architecture for Forecasting Currency Time Series

In this paper, we propose a singular spectrum analysis precoded deep learning architecture for forecasting a currency time series. The singular spectrum analysis is applied to extract the trend, fluctuation and noise components of the time series by means of the singular entropy. The deep learning architecture in the structure of the deep neural networks is used for model estimation and forecasting the components extracted from the time series. The performance in forecasting the THB/USD exchange rates in the currency exchange markets of the proposed singular spectrum analysis with deep neural networks architecture is provided and compared with that of the deep neural networks only.