Author(s) :

Adeev Mardia.

Volume/Issue :

Abstract :

This manuscript presents a comparative study of three machine learning models—Long Short-Term Memory (LSTM), Random Forest Regressor (RFR), and Support Vector Machine (SVM)—for predicting Bitcoin closing prices over the period from 2015 to 2023. The study focuses on evaluating the predictive performance of these models in terms of key metrics such as Mean Squared Error (MSE) and R-squared values, which measure their ability to forecast Bitcoin’s price trends. The findings reveal that the Random Forest Regressor (RFR) model outperforms both LSTM and SVM in terms of accuracy and robustness. RFR demonstrated the lowest MSE and the highest R-squared value, effectively capturing both short-term fluctuations and long-term trends in Bitcoin prices. LSTM exhibited moderate performance, struggling to capture extreme volatility, while SVM showed the poorest results, with the highest MSE and lowest R-squared value. In addition, this study explores the relationship between Bitcoin’s closing prices and trading volume, identifying a significant correlation that provides insights into market sentiment and price volatility. Challenges such as data preprocessing for SVM and hyperparameter tuning for LSTM are also discussed. The results underscore the potential of machine learning, particularly Random Forest, in enhancing cryptocurrency trading strategies and risk management. Future research will focus on integrating additional features and optimising models for real-time applications.