Título: Bitcasting: A Proposal of Multi -Horizon Bitcoin Direction Prediction Using Machine Learning
Autores: Hugo Leonardo Bezerra Fernandes, Daniel Carvalho da Cunha
Resumo: This paper investigates the efficacy of machine learning models in predicting Bitcoin’s price direction across multiple time horizons. We train models to classify price movements as buy, sell, or hold signals using technical, fundamental, and time-series features. Our methodology integrates feature engineering, data from Binance and CoinMetrics, and a comparative evaluation of XGBoost, support vector machine, logistic regression, and random forest. The results show that time series and technical indicators are especially effective at shorter horizons. Among the models, XGBoost consistently achieved superior performance in both classification accuracy and trading profitability. Simulated trading strategies based on the model’s predictions notably outperformed a buy-and-hold benchmark, reaching cumulative returns of 347% and 607% for 3-day and 30-day horizons, respectively, while maintaining competitive Sharpe ratios.
Palavras-chave: Bitcoin; cryptocurrency; machine learning; forecasting; time series analysis.
Páginas: 7
Código DOI: 10.21528/CBIC2025-1175712
Artigo em PDF: CBIC_2025_paper1175712.pdf
Arquivo BibTeX:
CBIC_2025_1175712.bib