Convolutional neural networks for forex time series forecasting

Abstract

The Deep learning approach plays a meaningful role in the prediction of financial time series data. A convolutional neural network (CNN) is a class of deep neural networks. The CNNs can automatically extract features and create informative representations of time series, eliminating manual feature engineering. This study aims to investigate the capability of 1D CNN to forecast time series. The multivariate multi-steps 1D CNN model is made and trained with the historical foreign exchange rate of EUR/USD. Intraday data in a 5-minutes time frame format are transformed into a three- dimensional structure to prepare the data for fitting a Convolutional Neural Network. Dataset preparation and CNN model are made using Python.