Predicting Bitcoin Log Returns: A Comparative Analysis of Machine Learning Models
Min Liu · Advances in Economics Management and Political Sciences · 2024
As the cryptocurrency market continues its rapid growth in both market value and trading volume, it has garnered significant attention. In response to this trend, this paper evaluates four machine learning models—Support Vector Regression (SVR), Random Forest (RF), Long Short-Term Memory (LSTM) networks, and Extreme Gradient Boosting (XGBoost)—to predict the log return of Bitcoin, one of the leading cryptocurrencies. The primary objective is to determine which model performs best in predicting Bitcoin's log return. During the research, the website Kaggle.com provides a solid database for all of the model-training processes and evaluations. Multiple experiments are then conducted to optimize each model's hyperparameters for maximum performance. Afterwards, in order to test whether models have learnt the features of data well, metrics like mean squared error (MSE), mean absolute error (MAE), and R-squared (R²) scores are computed and compared. According to the comparison, SVR outperforms the other models, achieving the best prediction accuracy. This paper offers some insightful information for analysts and researchers in the cryptocurrency domain. Additionally, it advances the area by offering a practical guide for utilizing machine learning algorithms to perform time series analysis, demonstrating their effectiveness in cryptocurrency market prediction tasks.