Optimization of Deep Learning Hyperparameters with Experimental Design in Exchange Rate Prediction

2020 61st International Scientific Conference on Information Technology and Management Science of Riga Technical University ( ITMS 2020): Proceedings 2020
Yunus Emre Mudilli, Sergejs Paršutins

Neural networks are widely used for exchange rate prediction. There are various hyperparameters affecting the prediction performance. In this paper, experimental design which is a statistical technique is used to identify the optimum hyperparameters within a given range for the prediction of USD/CAD exchange rate for 2019. In this context, multi-layer perceptron, recurrent neural networks and long-short term memories are considered as deep learning neural networks. The common hyperparameters for these neural network types are optimized via experimental design are epoch size, learning rate, batch size, number of hidden layer and number of hidden neurons. Optimum hyperparameter configurations are applied to predict test dataset.

Atslēgas vārdi
Deep learning, optimization, experimental design, exchange rate prediction
DOI
10.1109/ITMS51158.2020.9259300
Hipersaite
https://ieeexplore.ieee.org/document/9259300

Mudilli, Y., Paršutins, S. Optimization of Deep Learning Hyperparameters with Experimental Design in Exchange Rate Prediction. No: 2020 61st International Scientific Conference on Information Technology and Management Science of Riga Technical University ( ITMS 2020): Proceedings, Latvija, Rīga, 15.-16. oktobris, 2020. Piscataway: IEEE, 2020, 1.-4.lpp. ISBN 978-1-7281-9106-5. e-ISBN 978-1-7281-9105-8. Pieejams: doi:10.1109/ITMS51158.2020.9259300

Publikācijas valoda
English (en)