Πρόβλεψη της τιμής του Bitcoin με χρήση μηχανικής μάθησης, βαθιάς μάθησης και ανάλυσης συναισθήματος
Bitcoin price forecasting using machine learning, deep learning, and sentiment analysis
View/
Keywords
Bitcoin ; Deep Learning ; High-Frequency Data ; Machine Learning ; Price Forecasting ; Sentiment AnalysisAbstract
This postgraduate dissertation examines the prediction of Bitcoin prices over short-term forecasting horizons by employing statistical models, machine learning models, deep learning models, and sentiment analysis variables. The analysis is based on high-frequency BTC/USD data with one-minute intervals, covering the period from January 2022 to May 2026. The target variable is defined as the future logarithmic return for forecasting horizons of 1, 15, and 30 minutes, as logarithmic returns were considered more suitable than the absolute closing price due to the non-stationarity of the original price series. For the forecasting task, ARIMA, Ridge Regression, Random Forest, XGBoost, LSTM, and GRU models were implemented and compared, while model evaluation was conducted using RMSE, MAE, and Directional Accuracy, both at the level of logarithmic returns and predicted prices. In addition, textual data were collected and processed from Twitter, Reddit, and news sources, from which positive, negative, neutral, and compound sentiment variables were extracted using the VADER tool. These sentiment variables were temporally merged with the Bitcoin price data in order to examine whether external information from public discussion and news flow could improve the predictive performance of the models. The results showed that the inclusion of sentiment variables does not lead to a universal improvement across all models and forecasting horizons; rather, its effect varies depending on the model and the prediction horizon.
Postgraduate Studies Programme
Ψηφιακά Συστήματα και ΥπηρεσίεςDepartment
Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα Ψηφιακών ΣυστημάτωνNumber of pages
96Language
GreekCollections
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