A carregar...
Publicação
Forecasting Bitcoin Return Magnitudes Using Stacked Machine and Deep Learning Ensemble Modeling Approach
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 2.53 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Cryptocurrencies’ rapid growth and significant financial investments attracted traders seeking
to profit from their high volatility. As Bitcoin is the most valuable cryptocurrency by market
capitalization, innovative trading strategies using deep and machine learning are increasingly
being adopted to predict market dynamics, providing insights into trends and their drivers.
This study aims to develop a statistically robust and stochastically grounded stacked model by
combining novel machine learning and deep learning architectures to accurately forecast the
magnitude of price movements, specifically for the next day, using price action, technical
indicators and on-chain data. The CRISP-DM methodology was adopted to conduct this study.
Additionally, a novel feature selection architecture was developed to identify the most
predictive subset of input variables. This study evaluates the performance of a Random Forest
model combined with various neural network architectures explored to serve as base learners,
including a Recurrent Neural Network with Long Short-Term Memory layers (RNN-LSTM), a
Recurrent Neural Network with Gated Recurrent Unit layers (RNN-GRU), and a onedimensional Convolutional Neural Network (1D-CNN). The study further investigates the
effectiveness of several meta-models, namely Ridge Regression (RR), Multilayer Perceptron
(MLP), Extreme Gradient Boosting (XGBoost), Support Vector Regression (SVR), and Linear
Regression (LR). Among these, SVR, despite its simplicity, demonstrated strong statistical
performance. The RNN-LSTM model exhibited reliable generalization capabilities as a base
learner. Although XGBoost performed poorly in terms of traditional error metrics, it proved
valuable from a trading perspective, achieving a Mean Directional Accuracy (MDA) of 0.59 in
forecast instances where the absolute forecast exceeded the Mean Absolute Error.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Information Management, specialization in Business Intelligence
Palavras-chave
Bitcoin Forecasting Machine learning Deep learning Stacked models SDG 8 - Decent work and economic growth SDG 9 - Industry, innovation and infrastructure