Using Signal Decomposition Methods and Deep Learning Approaches to Forecast Bitcoin Price

Due to the outbreak of the COVID-19 epidemic in 2020, the price changes of virtual currencies are violent and irregular. Therefore, this research focuses on observing the prediction of Bitcoin price during the COVID-19 epidemic. In order to deal with the complex Bitcoin price change trend, this research predicts Bitcoin prices by combining signal decomposition methods with deep learning models. This research uses three signal decomposition methods—Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Variational Mode Decomposition (VMD), and Ensemble Patch Transform (EPT)—along with two composite signal decomposition methods—EPT-CEEMDAN and EPT-VMD—to decompose the original hourly Bitcoin price data from 2019 to 2023 into several sub-components, which are then individually input into deep learning models using Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (BiLSTM) and Temporal Convolutional Networks (TCN) for training and prediction. Finally, the predicted values of all sub-components are reconstructed to produce a final price prediction, and to compare the prediction error rates of various experimental combinations. Experimental results show that in most cases, combining signal decomposition methods can effectively improve model prediction performance, and the EPT-VMD-TCN method proposed in this research produced the best comprehensive prediction performance, with MAPE error rates of 0.0014, 0.004, 0.0007, 0.0076 and 0.0011 respectively for the 2019–2023 years.