Predicting Cryptocurrency Price Movements Using Leveraging Machine Learning Algorithms

Digital currencies are a new kind of innovation that is sweeping the cryptocurrency industry and the globe at large. The prices of cryptocurrencies are highly unpredictable, and receiving investments in these assets involve immense hazards and guesswork. This study uses machine learning techniques to predict future Bitcoin price changes by analyzing past data such as open, high, low, close, and transaction volume. After preprocessing the data to eliminate mistakes, duplication, and missing values, normalization is performed to guarantee consistency. There is a training set that contains <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 0 \%}$</tex> of the data and a testing set that contains 20%. The Adam optimizer and binary cross-entropy loss function are used to create a Gated Recurrent Unit (GRU) model. The model is trained across 200 epochs with a batch size of 32. The model performance is then evaluated with LSTM model with four metrics, which include <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{\mathbf{2}}$</tex>, MAE, RMSE and MSEs as well as XGBoost model. In regression tests, it is revealed that the proposed GRU model gives the best prediction performance with the foundation of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{\mathbf{2}}$</tex> 95.01%, a MAE of 16.31, RMSE of 20.91, and MSE of 43.74 against the LSTM model <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{\mathbf{2}} \mathbf{8 0. 6 1}$</tex> and XGBoost <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{\mathbf{2}} \mathbf{7 5}$</tex>. The GRU model's capacity to closely track real Bitcoin price movements, even minor variations during tumultuous times, is further supported by visualizations. Overall, the GRU model demonstrates superior performance in capturing temporal dependencies and provides the most reliable predictions among the evaluated models.