Unified Model Development for Predicting the Close-Price of Various Cryptocurrencies

Abstract

This study introduces a unified prediction model based on data fusion techniques for forecasting the prices of multiple cryptocurrencies. The cryptocurrency market is highly volatile and complex, characterized by dynamic interactions among various digital assets. Traditional models that focus on a single cryptocurrency often overlook these interconnections, limiting their predictive accuracy and practical applicability. To overcome this limitation, the proposed model integrates features from three major cryptocurrencies—Bitcoin (BTC), Ethereum (ETH), and Ripple (XRP)—and applies three data fusion strategies: Concatenation, Averaging, and a hybrid method that combines both Concatenation and Averaging. These feature fusion strategies are combined with four prediction algorithms: Support Vector Regression (SVR), Random Forest (RF), Gated Recurrent Unit (GRU), and Long Short-Term Memory (LSTM). To evaluate the model's performance and generalization capability, extensive prediction experiments were conducted on five cryptocurrencies: BTC, ETH, XRP, Monero (XMR), and Litecoin (LTC). The forecasting tasks were performed over four different time horizons: 5-day, 7-day, 24-day, and 30-day periods. The results demonstrate that for short-term forecasting (5-day and 7-day), the unified model using LSTM with average-based data fusion yields the highest prediction accuracy. For long-term forecasting (24-day and 30-day), the model employing GRU with average-based data fusion performs more effectively. In comparison with traditional single-cryptocurrency prediction models, the proposed unified model achieves notable performance improvements, with average reductions of 2.39% in Mean Absolute Error (MAE), 2.12% in Root Mean Square Error (RMSE), and 2.13% in Mean Absolute Percentage Error (MAPE). These improvements indicate that incorporating data fusion from multiple cryptocurrencies significantly enhances predictive performance. Overall, the findings validate the effectiveness and robustness of the unified data fusion approach in cryptocurrency price prediction. By capturing cross-asset relationships and leveraging advanced machine learning techniques, this study provides a more comprehensive and accurate prediction framework. This not only benefits investors seeking more reliable tools for decision-making but also contributes to the broader research on multi-source time series forecasting in highly volatile markets.

-