Long Short Term Memory (LSTM) model for Foreign Exchange Prediction

Abstract:

this work is focused on the attempt to predict Forex exchange rate using deep learning approach, thereby suggesting market price for Forex traders. There are several models or methodologies that can be used for deep learning like Convolutional Neural Network (CNN), and Restricted Boltzmann Machine (RBM). This work is directed towards concentrating on the Long Short Term Memory methodology in applying deep learning for Forex prediction. Datasets was collected, collated and organized from the Kaggle website, which contained Euro/US Dollar Forex (Foreign Exchange) trade results for the time period, from the 1st January 2011 to 31st December 2012.The Euro/US Dollar Forex dataset was tested with the Long Short Term Memory (LSTM) model and the outcomes for epoch value of 100 were obtained. These results were then plotted against the actual data gotten from the original dataset and it showed a close convergence in the predictions. This showed that the LSTM model can be used in making predictions, as the trends proved accurate enough even if not to the exact values, but proved to be very useful in making predictions. This will give Forex traders very useful tools in predicting forex market prices and enable them put together great decisions.

Keywords: Recurrent Neural Network (RNN), Long Short Term Memory (LSTM) model, Convolutional Neural Network (CNN), Restricted Boltzmann Machine (RBM)