Bilis Efstratios (Phd Candidate)

Thesis title: Creation of machine learning models - neural networks for the prediction of exchange rate values in different time frames

Supervisor: Goulianas Konstantinos

Advisory Committee Members:

Konstantinos Diamantaras, Professor, Dept. of Information and Electronic Engineering, IHU
Papadimitriou Theophilos, Department of Economics, DUT

Abstract:

My PhD research focuses on building machine learning models, particularly transformers, to
predict the next day’s close price for various currencies using time series data. The research
investigates the accuracy results of machine learning models applied in the forex field,
specifically focusing on currency pairs such as AUD/CAD, EUR/AUD, EUR/CAD, EUR/DKK,
EUR/JPY, GBP/AUD, NZD/USD, USD/CAD, USD/JPY, MYR/USD, USD/SGD, BRL/USD,
and NOK/USD. Various ML models, including Linear Regression, Random Forest, Stochastic
Gradient Descent (SGD), XGBoost (XGB), Long Short-Term Memory (LSTM), and
Transformers, have been developed for this purpose This entails collecting and preprocessing
relevant financial data, designing and training transformer-based models optimized for
forecasting, and evaluating their performance. The evaluation includes backtesting and
performance analysis using a comprehensive dataset spanning from 1999 to 2022. Metrics such
as Mean Absolute Error (MAE), Mean Squared Error (MSE), and Mean Squared Percentage
Error (MAPE) are utilized to assess the models’ effectiveness in capturing and predicting the
next day’s close price.