Bitcoin price prediction using machine learning

• Main Author: Tang, Jian Yang
• Format: Final Year Project / Dissertation / Thesis
• Published: 2023
• Subjects: T Technology (General); TA Engineering (General). Civil engineering (General)
• Online Access: http://eprints.utar.edu.my/6045/1/fyp__CS_2023_TJY.pdf; http://eprints.utar.edu.my/6045/

Bitcoin price prediction is the act of forecasting future value of Bitcoin. A successful prediction of Bitcoin future value will maximize investor’s gains. Over the past few years, Bitcoin has been a topic of many, from investors to researchers, even ordinary citizens. Bitcoin is the first, largest and most valuable cryptocurrency till today. However, Bitcoin’s nature is very volatile and highly fluctuate which makes investing in Bitcoin feels like gambling to investors. It is very risky to invest in it as its price go up and down a lot within 1 day interval. Numerous studies have conducted on Bitcoin price prediction using traditional time series forecasting algorithms. In recent years, researchers have started using deep learning models to predict Bitcoin price as well. This study proposes three types of machine learning algorithms (LSTM, GRU, and Prophet) with two types of architectural configurations (Sequence-to-Sequence and Sequence-to-One) to predict Bitcoin’s closing price based on 1 year of Bitcoin historical data, (2, April 2022 to 2, April 2023). The data is split into 335 days for training set and 30 days for testing set. Three experiment was conducted (Sequence-to-Sequence Walk Forward, Sequence-to-One Walk Forward, and Sequence-to-One Rolling Origin). The performance of the proposed models is evaluated using Bitcoin price from 3, March, 2023 to 1, April 2023. The results on the models using various evaluation metrics such as RMSE, MAPE and MAE show that LSTM is the optimal model compared to GRU and Prophet. GRU is a second close. However, Prophet struggles to predict fluctuations and curvature in this highly unstable Bitcoin price prediction task. Having lower error metrics does not imply that the model is good.