Bitcoin Price Trends: A Neural Network Approach with RNN & LSTM

The present study presents a new method for predicting Bitcoin price in USD using deep neural network techniques on blockchain data. The proposed model applies a hybrid framework being an integration of Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) networks, utilizing RNNs to capture short-term dependency and LSTMs to embed long-term structures within the evolution of prices. Historical price data from the Bitcoin Price Index is used, and blockchain-specific features—transaction volume, mining difficulty, and network congestion are added to augment the predictive feature set. Bayesian optimization is applied for hyperparameter tuning to maximize accuracy and efficiency. The model achieves a classification accuracy of 52% and an RMSE of 8%, outperforming traditional statistical methods like the ARIMA model in forecasting Bitcoin's non-linear price action. In addition, parallelization according to GPU accelerates model training. The research findings show the efficiency of hybrid deep learning techniques in forecasting cryptocurrency price trends, providing valuable information to investors, traders, and financial analysts.