Using Machine and Deep Learning Models, On-chain Data, and Technical Analysis for Predicting Bitcoin Price Direction and Magnitude

Abstract

This study investigates the performance of Machine Learning (ML) and Deep Learning (DL) models in predicting both Bitcoin price magnitude and movements using a unique combination of feature selection, input types, and output types. The model inputs are Bitcoin price, on-chain metrics, and technical analysis indicators (TA). The model outputs are Bitcoin price direction and magnitude. The ML and DL models developed in this paper include Support Vector Machines (SVM), Random Forest, Gradient Boosting Machines, Long Short-Term Memory, Convolutional Neural Network-Long Short-Term Memory, Gated Recurrent Units, Temporal Convolutional Network, and Long and Short-Term Time-series Network. An important aspect of this study was the use of the Boruta feature selection method to select the most relevant subset of features. When predicting price movement, the findings revealed that the SVM model achieves the best performance with an accuracy of 83% and F1-Score of 82%. When predicting price magnitude, the model achieved the lowest RMSE (1,531.3) and an impressive R2 (0.9856). Generally, the DL models gave competitive performance for the classification tasks, but they mostly performed poorly with regression tasks. It was discovered that the inclusion of the Boruta feature selection method significantly improved performance of the models. On-chain data also significantly improved model performance, especially for classification tasks. Finally, the study assessed the models' profitability through backtesting, and Boruta-SVM emerged as the most profitable model. The classification models mostly obtained positive returns, while the regression models yielded losses or marginal gains. This study emphasizes the importance of feature selection, model choice, and target type in Bitcoin price prediction. Analysts can use these insights to improve investment decisions, thereby maximizing returns.

Department(s)

Engineering Management and Systems Engineering

Keywords and Phrases

Back testing; Bitcoin; Cryptocurrency; Deep learning; Feature selection; Machine learning

International Standard Serial Number (ISSN)

0952-1976

Document Type

Article - Journal

Document Version

Citation

File Type

text

Language(s)

English

Rights

© 2025 Elsevier; International Federation of Automatic Control (IFAC), All rights reserved.

Publication Date

15 Aug 2025

Link to Full Text

Share

 
COinS