Bitcoin Price Forecasting using Blockchain Metrics, Ensemble Machine Learning Algorithms and Facebook Prophet
Year: 2024
Type: article
Abstract: As one of the most well-known cryptocurrencies, Bitcoin has gradually been embraced, showcasing the potential to emerge as a common method of payment. The year 2021 witnessed considerable fluctuations in Bitcoin’s value, sparking a vast amount of research published in the Web of Science since the first studies on Bitcoin and cryptocurrencies appeared in 2012. The volatility of 2021, marked by successive bear and bull markets, highlighted the limitations of existing prediction tools. The main concern with Bitcoin lies not in its volatility, but rather in its unpredictability. Effective forecasting is crucial as it enhances retailers’ strategies towards Bitcoin, encouraging its broader adoption. This study aims to conduct mid-term price predictions with a five-day outlook, utilizing Bitcoin previous prices, blockchain metrics and Twitter sentiments as input data. The prediction technique employs a combination of ensemble machine learning methods, including bagging, boosting, voting, and stacking, to achieve its forecasts. The analysis covers daily data collected from January 1, 2019, to May 31, 2022. For longer-term forecasts, the Facebook Prophet tool is used, suitable for univariate data sets only, demonstrating commendable accuracy for long-term projections. The Mean Absolute Error (MAE) metric is used to evaluate the precision of these predictions.
Author: Simona-Vasilica Oprea
Institution: Bucharest University of Economic Studies
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: diamond
Award: G6694228673
OpenAlex ID: W4410551009