Abstract:
Predicting Bitcoin prices remains a complex challenge due to the cryptocurrency
market's inherent volatility and rapid fluctuations. Based on historical price data
gathered from Investing.com over the previous five years, this study investigates the
potency of three sophisticated deep learning models for predicting Bitcoin prices:
Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), and Bidirectional
Long Short-Term Memory (Bi-LSTM). Rigid backtesting and comparative analysis
were used to assess each model's ability to predict future Bitcoin values. With an
accuracy of 98.02%, a Mean Absolute Error (MAE) of 1023.48, a Root Mean Square
Error (RMSE) of 1576.11, and an R2 score of 0.9904, the findings show that the GRU
model performs better than the LSTM and Bi-LSTM models. An accuracy of 97.99%
was attained by the LSTM model with an MAE of 961.70, an RMSE of 1426.48, and
an R2 score of 0.9921; on the other hand, a 96.91% accuracy was attained by the
Bi-LSTM model with an MAE of 1441.28, an RMSE of 1848.11, and an R2 score of
0.9867. This study shows that the GRU model performs better than the other models
and demonstrates the usefulness of deep learning methods for predicting Bitcoin
prices. Additionally included in the paper are the effects of these models on market
volatility, ethical issues in cryptocurrency trading, and financial decision-making. The
results provide the groundwork for future lines of inquiry that may include outside
influences and create real-time prediction systems. In summary, this study advances
financial technology by showing that complex neural network architectures can
accurately predict Bitcoin prices and provide information about potential future
research directions for enhancing predictive models and tackling ethical and
environmental issues in the cryptocurrency space.