Time-Series Forecasting of Ethereum Price using Long Short-Term Memory (LSTM) Networks

Abstract

In recent times, ether (ETH) has become one of the most popular cryptocurrencies that is gaining significant interest from crypto investors and developers across the globe. The increased interest in this cryptocurrency is due to the fact that transac tions on the Ethereum platform are far more secure, as it combines smart contracts to streamline commerce and trade between both anonymous and recognized parties. Besides, many decentralized financial and nonfinancial apps (DeFi and DApps) are built mainly based on the ether cryptocurrency itself. As a result, the price of this cryptocurrency is also rising gradually. On the other hand, the price of ether some times decreases as well due to some unwanted circumstances like political conflicts, wars, natural disasters, and so on. Thus, the ether cryptocurrency market has be come very unpredictable and can cause an uncertain situation for market investors. For this purpose, having a specialized prediction method for the ether price based on machine learning and deep learning technologies is crucial. This research aims to find an accurate price prediction model for the ether cryptocurrency based on the long short-term memory (LSTM) network, which is a special variant of the recur rent neural network (RNN). In the proposed model, ether price data was taken in time-series format and fitted into multiple basic and hybrid variants of the LSTM network, and the future prices were predicted based on both univariate and mul tivariate time-series analysis. Furthermore, a comparative analysis was conducted among the models and also some popular existing forecasting techniques like autore gressive integrated moving average (ARIMA) as the baseline forecast to determine which one can provide the best possible accuracy so that investors may understand the behaviour of the ether market and make proper decisions on their investment.