NEAT vs LSTM vs XGBoost. Three novel methods introduced and compared on forex trading
View/
Keywords
Neuro Evolution of Augmenting Topographies (NEAT) ; FX market ; Foreign exchange market ; Time series ; LSTM ; Currency ; Trading ; XGBoost ; Deep learningAbstract
Time series forecasting can be very challenging in financial markets especially in cases like the forex (FX) markets. Its complexity, led many researchers in the forecast of the direction of trade rather than the actual value, a measure that can be priceless for traders and strategists. On the other hand, since its introduction, the Neuro Evolution of Augmenting Topologies (NEAT), is a genetic evolutionary algorithm that has not been extensively used and tested in this type of data while it seems to have obvious advantages compared to other approaches. In reviewed literature, the Long Short-Term Memory (LSTM) is highlighted as one of the most efficient deep learning methods for dealing with time series forecast regression problems. Therefore, the focus of this paper is to introduce a novel application of NEAT and compare with an LSTM model to predict the direction of the EUR/USD. Moreover, these are compared with an XGBoost machine learning application that in recent literature seems to function well with stock market data. The results suggested, that all the proposed approaches can be very effective in the forecast of the direction of currency market. This research is indicative of the potential that advanced algorithms have in dealing with complex tasks like the prediction of FX.
Postgraduate Studies Programme
Προηγμένα Συστήματα ΠληροφορικήςDepartment
Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα Ψηφιακών ΣυστημάτωνNumber of pages
80Language
EnglishCollections
Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα