Case Study: Predicting Future Forex Prices Using MLP and LSTM Models

Abstract

This study compiles 10 years of past daily closing prices for numerous trading assets, and economic reports. The data will be used to train, test, and validate a variety of ANN and LSTM models. This paper introduces a learning window that trains the network using a predefined number of days in the past to predict several days' closing prices in the future. A search grid algorithm is used to select the optimal hyperparameters for the networks. The MLP network managed to achieve an accuracy of above 80% when 30 days of the previous closing price were used as input to the network.