COMPARATIVE ANALYSIS OF SHORT-TERM AND LONG-TERM FORECASTING PERFORMANCE: ARIMA VS. LSTM MODELS IN TIME SERIES PREDICTION

Abstract

This study conducts a comparative analysis of the ARIMA and LSTM models for forecasting Bitcoin prices, focusing on key performance metrics such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and overall accuracy. The ARIMA model, while achieving a commendable accuracy of 98.21%, exhibited higher error rates with an MAE of $837.77 and an RMSE of $940.40, reflecting its limitations in handling long-term and highly volatile data. Conversely, the LSTM model demonstrated superior performance with a remarkable accuracy of 99.73%, an MAE of $126.97, and an RMSE of $151.95, showcasing its ability to capture non-linear dependencies and complex temporal patterns inherent in Bitcoin price data. The results indicate that LSTM models are more suitable for dynamic and intricate time series datasets, while ARIMA models remain effective for short-term forecasting with linear trends. The study concludes with recommendations for leveraging the strengths of both models, emphasizing the use of LSTM for complex forecasting tasks and suggesting hybrid approaches to enhance predictive accuracy.