Research on Bitcoin price prediction based on Support Vector Regression and its variant combination model

Abstract

Bitcoin is one of the most successful cryptocurrencies, and research on Bitcoin price prediction is getting more and more attention. Previous studies have used traditional statistical methods and machine learning models to predict Bitcoin prices. However, previous studies also have many problems, such as too few influencing factors, lack of model optimization, and poor prediction effect. This paper selects 27 factors related to Bitcoin price changes and screens the features through the XGBoost algorithm and the Random Forest algorithm (RF). In this study, combined forecasting models based on Support Vector Regression (SVR), Least Squares Support Vector Regression (LSSVR) and Twin Support Vector Regression (TWSVR) are used to predict Bitcoin price, separately. In addition, the Whale Optimization Algorithm (WOA) and Particle Swarm Optimization (PSO) are applied for parameter tuning of the models. Expected Variance Score (EVS), Coefficient of Determination (R2), Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE) are used to measure the prediction accuracy of the combined models. The CPU time is used to measure the operation speed of the combined models. The experimental results show that the combined model XGBoost-WOA-TWSVR has the best prediction effect, and the EVS score of this model is 0.9547. In addition, our research verifies that Twin Support Vector Regression has advantages in both prediction effect and computational speed.