Foreign exchange prediction using long short-term memory neural network
Sim, Ming Shi
Date of Issue2019
School of Electrical and Electronic Engineering
Long short-term memory (LSTM) neural networks are a modern machine learning technique for sequence learning and prediction. They are inherently suitable and commonly applied to financial time series prediction problems. In this paper, the Author introduces four multivariate models based on LSTM neural networks to forecast foreign exchange (Forex) rates comprising Euro against US Dollar (EUR/USD), US Dollar against Japanese Yen (USD/JPY), British Pound Sterling against US Dollar (GBP/USD), and US Dollar against Swiss Franc (USD/CHF). The Author examines hyperparameters including number of hidden layers and hidden neurons, number of epochs and batch size, dropout rate, and sliding window width and finds them to be key determinants of the performance of a trained neural network. Experimental results in comparison with Artificial Neural Network (ANN), Convolutional Neural Network (CNN), Support Vector Regression (SVR), and Gated Recurrent Unit (GRU) illustrate the effectiveness of the tuned LSTM models in Forex predictions.
DRNTU::Engineering::Electrical and electronic engineering
Final Year Project (FYP)
Nanyang Technological University