DefinePK

Abstract

The highly stochastic, nonlinear, and volatile nature of Bitcoin prices poses significant challenges for accurate forecasting using traditional statistical models. To address this, we propose a hybrid deep learning architecture that combines the strengths of Convolutional Neural Networks (CNNs) for spatial feature extraction with Recurrent Neural Networks (RNNs), specifically Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU), for capturing long-term temporal dependencies. This integrated framework effectively models both spatial and temporal patterns from historical Bitcoin price data. The model was trained and evaluated using real-world Bitcoin datasets.Experimental results demonstrate that the proposed CNN+LSTM model outperforms traditional machine learning and standalone deep learning approaches. Specifically, it achieves a Root Mean Square Error (RMSE) of 245.76, a Mean Absolute Error (MAE) of 11.45, a Mean Absolute Percentage Error (MAPE) of 15.68%, an R² score of 0.92, and a Mean Bias Error (MBE) of 6.94. These results highlight the effectiveness and reliability of the proposed hybrid model in enhancing the accuracy and stability of financial time series forecasting, providing valuable insights for traders, investors, and financial analysts.