Μοντελοποίηση αγοράς πλοίων μεταφοράς φορτίου χύδην με χρήση αναδρομικών νευρωνικών δικτύων

Σαραντόπουλος, Φώτιος; Sarantopoulos, Fotios

dc.contributor.author	Σαραντόπουλος, Φώτιος	el
dc.contributor.author	Sarantopoulos, Fotios	en
dc.date.accessioned	2022-01-18T20:08:21Z
dc.date.available	2022-01-18T20:08:21Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/54361
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.22059
dc.rights	Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα	*
dc.rights.uri	http://creativecommons.org/licenses/by-nc-nd/3.0/gr/	*
dc.subject	Recurrent neural networks	en
dc.subject	Machine learning	en
dc.subject	Baltic Dry Index	en
dc.subject	LSTM	en
dc.subject	GRU	en
dc.title	Μοντελοποίηση αγοράς πλοίων μεταφοράς φορτίου χύδην με χρήση αναδρομικών νευρωνικών δικτύων	el
dc.title	Modelling the Dry Bulk Shipping Market with the use of Recurrent Neural Networks	en
heal.type	bachelorThesis
heal.classification	Μηχανική μάθηση	el
heal.classification	Ναυπηγική	el
heal.classification	Νευρωνικά δίκτυα	el
heal.classification	Neural Networks	el
heal.language	en
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2021-10-21
heal.abstract	The shipping industry is an area that is subject to changes and strong fluctuations. The unstable nature of freights makes the process of predicting their value a very demanding problem that depends on multiple factors and the solution of which has the potential to provide significant financial profit margins. In this thesis, we will focus on the regression problem of predicting the price of the Baltic Dry Index with the use of several models, proving that Recurrent Neural Networks (LSTM, GRU, BiLSTM) are very efficient in time-series predictions. Firstly, the basic concepts and the operation of the shipping market are presented, as well as the macroeconomy in general. Then, the theoretical background of neural networks is analysed in order to make them more understandable to the reader. Based on the operation of the shipping market, and more specifically the dry bulk shipping market, shipping and macroeconomic features are examined by using a set of feature selection techniques such as the feature correlation and the feature importance techniques. In addition, we develop a set of machine learning models: LSTM, GRU, BiLSTM, Feedforward and ARIMA in order to predict the price of BDI in a time window of one month, which is a regression problem. The recurrent neural networks (LSTM, GRU and BiLSTM) have the best performance in the regression problem for a period of one month, with the LSTM being the most accurate and providing the lowest errors. In addition, the ARIMA statistical model showed a high accuracy as well, while the feedforward neural networks provided high error rates and not accurate results. The results of this thesis confirm that the use of macroeconomic variables, other than the shipping ones, are very beneficial for predictions in the shipping market. Finally, they also establish the fact that Recurrent Neural Networks are precise decision-making tools, and can be applied to a large set of predictions of timeseries, highlighting their ability to predict prices with great accuracy.	en
heal.advisorName	Lyridis, Dimitrios	en
heal.committeeMemberName	Tolis, Athanasios	en
heal.committeeMemberName	Ventikos, Nikolaos	en
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Ναυπηγών Μηχανολόγων Μηχανικών. Τομέας Μελέτης Πλοίου και Θαλάσσιων Μεταφορών	el
heal.academicPublisherID	ntua
heal.numberOfPages	125 σ.	el
heal.fullTextAvailability	false