Comparison of State-of-art Machine Learning Models’ Performances: Evidence from BTC, BNB, ETH, USDC and USDT
Year: 2024
Type: article
Abstract: As a matter of fact, cryptocurrency price prediction has become a major trend for financial research especially based on the state-of-art machine learning techniques. This study explores the effectiveness of six machine learning (ML) models in predicting the daily prices of five major cryptocurrencies, i.e., Bitcoin (BTC), Ethereum (ETH), Tether (USDT), BNB (BNB), and USD Coin (USDC). The ML models applied include the Ordinary Least Squares (OLS) model, Random Forest, Long Short-Term Memory (LSTM), XGBoost, LightGBM, and Transformer. The results demonstrate that traditional regression and tree-based ensemble models, particularly OLS and LightGBM, are more effective in predicting cryptocurrency prices, especially for BTC, BNB, and ETH. LightGBM consistently outperforms OLS in minimizing prediction errors, while Random Forest and XGBoost also yield strong, though slightly less accurate, results. Conversely, LSTM and Transformer models underperform, indicating their limitations with the specific characteristics of cryptocurrency data. These findings suggest that traditional and tree-based models are better suited for cryptocurrency price prediction. Future research should investigate the incorporation of additional market indicators and advanced ML architectures to enhance predictive accuracy.
Author: Zimeng Yi
Institution: Nanyang Technological University
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W4405820818