ARIMA for short-term and LSTM for long-term in daily Bitcoin price prediction

Abstract

The goal of this paper is the insight into the forecasting of Bitcoin price using machine learning models like AutoRegressive Integrated Moving Average (ARIMA), Support vector machines (SVM), hybrid ARIMA-SVM, and Long short-term memory (LSTM). Depending on the different types of data and the period, various models are used for prediction. A single model may be the best fit in the short term but may not be the best in long-term series data. Thus, using only a single model may not be suitable for forecasting time series data that depends on data sampling length and prediction time, and the type of specific applications. As a result, the ARIMA model produces better error results with a short prediction period or a small data set. In contrast, the Hybrid ARIMA-SVM model will help improve the performance of the ARIMA model when predicting over a long period, specifically 7 and 30 days for Bitcoin price prediction used in this research paper. The paper aims to compare traditional models such as the ARIMA, the Hybrid ARIMA-SVM, and deep learning models such as LSTM on a specific cryptocurrency prediction task using different scenarios.