Comparative Analysis of GRU and Bi-LSTM Models for Bitcoin Price Prediction

Bitcoin price prediction is a challenging yet crucial task for traders, investors, and financial analysts due to the cryptocurrency’s inherent volatility and sensitivity to global economic factors. Reliable forecasting models can support for informed decision-making in trading strategies, risk management, and market analysis. This study explores the application of deep learning models as gated recurrent units (GRU), long short-term memory (LSTM), and bidirectional LSTM (Bi-LSTM) to forecast Bitcoin closing prices using historical market data. The dataset used consists of historical Bitcoin price data, including daily closing prices, trading volumes, and other key financial indicators. Comprehensive data pre-processing techniques were applied, including handling of missing value imputation, normalization, and feature engineering to uncover relevant patterns. Model performance was evaluated using metrics such as mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), R-squared (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>), and variance scores, offering a robust framework for comparing predictive performance. Our findings reveal that while the Bi-LSTM model achieved the highest training R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>-score, it exhibited the lowest test R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>-score, indicating signs of overfitting. In contrast, the GRU model demonstrated superior test performance, attaining the highest test R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>-score, and the lowest error values across RMSE, MSE, and MAE.