Author(s)

Siyu Yao, Di Ma, Yingying Zhang

Corresponding Author

Siyu Yao

ABSTRACT

Bitcoin is the world's leading cryptocurrency and allows users to conduct anonymous transactions safely over the Internet. Bitcoin has attracted the attention of investors in recent years. We obtained the historical data of Bitcoin prices from 2010 to 2020 and analyzed the trend of Bitcoin prices based on 10 technical indicators. Then several machine learning methods such as the Logistic regression, Support Vector Machine (SVM), Random Forest (RF), XGBoost and lightGBM were used to predict movements of Bitcoin Prices. The experimental results showed that lightGBM was the most accurate in predicting the movements. In addition, we found that the results calculated with discrete variables of technical indicators were better than that with continuous features as input. Principal Component Analysis could reduce dimension so that the prediction performance of all machine learning models was improved. Finally, we integrated three models SVM with Gaussian kernel, XGBoost and lightGBM by the idea of Bagging. The total return reached about 1200% using Bagging method to buy or sell Bitcoin in a year and a half, which is prior to lightGBM method. Machine learning methods and the fusion model used for predicting Bitcoin price could also solve the problem of other digital currency price prediction.

KEYWORDS

Technical indicators, Bitcoin price movements, Machine learning, Model fusion