Προβλέψεις χρονολογικών σειρών μέσω τεχνητών νευρωνικών δικτύων
Neural networks for time-series forecasting
View/ Open
Keywords
Νευρωνικά δίκτυα ; MLP ; CNN ; ΧρονοσειρέςAbstract
The ability to make predictions based on historical time series data is highly significant and has applications in various fields such as econometrics, finance, environmental science, biology, telecommunications, and many others. The purpose of this MSc Thesis is to provide a detailed presentation of a wide range of time series prediction methods based on feed-forward artificial neural networks. More specifically, algorithms for developing Convolutional Neural Networks and Multilayer Perceptrons will be described and implemented. These networks will be trained using historical data from the QQQ ETF time series by Invesco, which aims to replicate the performance of the Nasdaq-100 index. Additionally, we will leverage some of the most significant stocks that contribute to the formation of this index. The performance of the model predictions will be empirically examined and compared to traditional forecasting methods such as the ARIMA model.