Forecasting Bitcoin Price Using LSTM Networks Optimized with Adam Optimizer

Abstract

The rapid growth of cryptocurrency markets has increased the importance of accurate price forecasting to support investment decision-making and risk management. Bitcoin, as the most dominant cryptocurrency, exhibits high volatility, nonlinearity, and strong temporal dependencies, which make its price dynamics difficult to model using traditional statistical approaches. Motivated by these challenges, this study proposes a Bitcoin price forecasting model based on a Long Short-Term Memory (LSTM) neural network to capture complex temporal patterns in historical price data. The main contribution of this research lies in the development and evaluation of an LSTM-based forecasting framework that effectively models long-term dependencies in Bitcoin price movements. The proposed methodology includes data preprocessing, normalization, sequence transformation, model training, and systematic performance evaluation. The forecasting performance is assessed using standard regression metrics, namely Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE), to ensure objective and comprehensive evaluation. Experimental results demonstrate that the proposed LSTM model is capable of producing predictions that closely follow actual Bitcoin price trends during the testing period, indicating strong predictive accuracy and robustness despite market volatility. The findings confirm the suitability of LSTM networks for cryptocurrency price forecasting tasks involving nonlinear and non-stationary time series data. Future work may extend this research by incorporating external factors such as trading volume, market sentiment, and macroeconomic indicators, as well as by exploring alternative deep learning architectures and multi-step forecasting strategies to further enhance prediction performance.