Conclusion

The Yo-Yo 10-minute prediction model demonstrates the effectiveness of the Decomposed LSTM (DLSTM) architecture in capturing complex temporal dependencies in cryptocurrency markets. Through the implementation of time-decomposition mechanisms that enhance traditional LSTM capabilities, the model has shown robust performance across varying market conditions throughout the 2020-2024 testing period. The entropy-based confidence filtering enables dynamic adaptation to market volatility, with prediction accuracy consistently exceeding 75% when filtered for high-confidence signals (>50%). These results validate both our classification framework and confidence measurement approach, establishing a strong foundation for high-frequency cryptocurrency price movement prediction at the 10-minute horizon, and providing a useful signal for trading.