Comparison of Machine Learning Models of Long Short-Term Memory and Facebook Prophet Model in Ethereum Gas Fee Determinant

Abstract

Blockchain technology is one of the most novel technologies that received attention from academia and practitioners in various industries because of its profound nature and the opportunities that it offers, especially in the digital age. Ethereum, which is the largest decentralized blockchain software, was introduced in 2015 and is best known for its smart contracts that facilitate different utilizations over its blockchain network. Also, it can be used as a cryptocurrency, and historically it has been the 2nd most valuable cryptocurrency following Bitcoin. In order to be able to use Ethereum smart contracts over the blockchain network, a transaction fee or a gas fee is set by the network according to that specific task. This study is aimed to develop Machine Learning (ML) models to predict the Ethereum gas fee by using a comparison of Long Short-Term Memory (LSTM) and the Facebook Prophet Model (FPM). The gas fee prediction modeling was done based on the daily dataset of eight years from 2015 to 2022. The results in this study showed that the FPM model in the scenario of this study resulted in a Mean Absolute Error (MAE) of 0.02 and a Root Mean Squared Error (RMSE) of 0.05. However, MAE and RMSE values of the LSTM model turned out equal to 0.006 which showed a higher performance and more accuracy compared to the FPM model.