Comparative Performance of LSTM and ARIMA for the Short-Term Predictive Analysis

The rapid growth of digital assets, particularly Bitcoin, has introduced extreme price volatility that challenges investors and forecasters. Accurate prediction models are crucial to reduce financial risks and support data-driven investment decisions. This study aims to compare the performance of two forecasting approaches—Autoregressive Integrated Moving Average (ARIMA) and Long Short-Term Memory (LSTM)—using daily Bitcoin price data from 2022 to 2025. Both models were implemented under the same experimental framework and evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). The results show that LSTM outperforms ARIMA across all metrics (MAE: 7,235.77; RMSE: 8,202.41; MAPE: 6.43%), demonstrating better adaptability to nonlinear and volatile price movements. Despite longer training time, LSTM provides more stable and precise forecasts. The findings conclude that deep learning models are more effective for cryptocurrency forecasting, and hybrid combinations can enhance predictive robustness