Short term return prediction of cryptocurrency based on XGBoost algorithm

Abstract

The price of cryptocurrency is easily affected by various economic, political and other factors, with huge fluctuation, which makes it difficult to predict, compared with stocks and other financial products. Therefore, the prediction of its short-term return in this paper can provide some valuable suggestions for investors. This paper uses XGBoost algorithm to predict 14 kinds of cryptocurrency markets, experiments based on the data applied by KAGGLE competition platform, and expands the data features combined with feature engineering. Experimental data express that our advanced model has significantly improved forecast performance compared with other traditional machine learning algorithms. Specifically, the prediction performance of XGBoost algorithm is 12.5%, 16.6% and 43.3% higher than that of Gradient Boosting model, SVM algorithm and Linear Regression algorithm respectively. In addition, we also rank the importance of all the features of the simulation, and give some constructive suggestions to guide the future work.