Forecasting Bitcoin Prices in the Context of the COVID-19 Pandemic Using Machine Learning Approaches
Conference paper
Sontakke, P., Jafari, F., Saeedi, M. and Amirhosseini, M. 2023. Forecasting Bitcoin Prices in the Context of the COVID-19 Pandemic Using Machine Learning Approaches. 4th International Conference on Data Analytics & Management (ICDAM-2023). London, UK 23 - 24 Jun 2023 Springer.
Authors | Sontakke, P., Jafari, F., Saeedi, M. and Amirhosseini, M. |
---|---|
Type | Conference paper |
Abstract | Using daily data from April 1st, 2016 to March 3rd, 2022, this study aims to explore the use and effectiveness of machine learning algorithms in forecast-ing the price of Bitcoin. The paper examines the forecasting performance based on different time lags within the selected periods: 1) Before Pandemic and 2) Including Pandemic. The second time frame is selected to examine the effect of the Covid pandemic on the Bitcoin market fluctuations. This research employs four machine learning models, including Linear Regression, Support Vector Regression, Extreme Gradient Boosting, and Long Short-Term Memory. These are refined and calibrated to produce the most accurate forecasts. The performance of the algorithms was measured and compared using regression metrics. The results show that before the pandemic, the linear regression model performed the best for next-day predictions, while Ex-treme Gradient Boosting performed best overall and for longer-term predictions. For the period including the pandemic, Extreme Gradient Boosting and linear regression performed the best, consistently outperforming Long Short-Term Memory and Support Vector Regression. The prediction models for da-ta before the pandemic have demonstrated improved performance, whereas the selected model for the period including the pandemic exhibited satisfactory results. This is because bitcoin prices displayed the highest volatility during the Covid pandemic. The study finds that Extreme Gradient Boosting performs best overall and for longer-term predictions, while linear regression performs the best for next-day predictions before the pandemic. Moreover, the study reports satisfactory results for bitcoin price prediction for the period including the pandemic, despite the high volatility of prices. |
Keywords | Cryptocurrency; Bitcoin Price; Time Series Forecasting; Machine Learning; Technical Indicators; Linear Regression; Support Vector Regression; Extreme Gradient Boosting; Long Short-Term Memory |
Year | 2023 |
Conference | 4th International Conference on Data Analytics & Management (ICDAM-2023) |
Publisher | Springer |
Accepted author manuscript | License File Access Level Repository staff only |
Publication process dates | |
Accepted | 17 Mar 2023 |
Deposited | 09 May 2023 |
Web address (URL) | https://www.icdam-conf.com/ |
Copyright holder | © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG 2022 |
https://repository.uel.ac.uk/item/8vz2w
Restricted files
Accepted author manuscript
14
total views7
total downloads1
views this month0
downloads this month