Comparative Analysis of LSTM, GRU, and Bi-LSTM Deep Learning Models for Time Series Cryptocurrency Price Forecasting

Cryptocurrency is a highly volatile digital asset that requires accurate predictive methods. This study compares the performance of three deep learning architectures LSTM, GRU, and Bi-LSTM in forecasting the prices of Bitcoin (BTC), Ethereum (ETH), and Binance Coin (BNB) using univariate historical data. Evaluation was conducted through regression metrics (RMSE and MAPE) and classification of price movement into five categories, ranging from very bearish to very bullish, assessed using a confusion matrix. The results show that GRU performed best for BTC (RMSE 974.72, MAPE 1.18%), while Bi-LSTM outperformed others for ETH and BNB (RMSE 43.19 and 6.83; MAPE 1.16% and 1.08%) and achieved the highest classification accuracy (55% and 52%). However, overall classification accuracy remains low, reflecting the complexity of cryptocurrency price patterns. The study is limited by its univariate approach without incorporating external variables. Its contribution lies in combining regression and classification evaluation, and it recommends exploring multivariate and ensemble models in future research.