Νευρωνικά δίκτυα και εφαρμογές αυτών για την πρόβλεψη χρονοσειρών στο χρηματιστήριο
View/
Keywords
Νευρωνικά δίκτυα ; Πρόβλεψη χρονοσειρών ; MATLABAbstract
In the context of this thesis, the problem that was examined concerned the prediction of time series in the stock market using neural networks. From a bibliographic point of view, introductory concepts on neural networks and time series prediction methods were initially presented. As for the practical part of the thesis, it was decided to test three different neural network topologies, and experimental results were obtained for various prediction time windows. The evaluation of the three different architectures was carried out in terms of the RMSE metric, but also in terms of the use of graphs to compare predicted and actual values. It is also worth mentioning that the experimental procedure took place in a MATLAB environment and all the input data came from the S & P 500 stock index. A first conclusion, derived from carrying out the experimental procedure is that in all 3 cases of architectures and for all time window values satisfactory results were obtained in terms of the error between actual and predicted values. A second conclusion is that the large volume of the dataset (records covering the years 1950 to 2014), as well as the normalization of the data implemented, contributed in a positive way to the increase in the quality of the results .In addition, the MATLAB environment has been a quite useful tool for carrying out the experimental process. Important advantages are the user-friendly interface, as well as the use of ready-made functions for the creation of each neural network. In addition, the easy creation of graphs played an important role in the analysis and visualization of data.