Time Series Analysis by Bi-GRU for Forecasting Bitcoin Trends based on Sentiment Analysis

Abstract

In the last few years, the cryptocurrency market, especially Bitcoin, has attracted many people, including machine learning engineers. They have heavy competition in predicting the price or the rise and fall of the price in the future. To achieve this goal, they used various types of approaches like Linear Regression, SVM, and deep learning methods like Neural networks, Recurrent neural networks like RNN, LSTM, and GRU, Bidirectional neural networks, and a combination of methods. M.L and D.L engineers used various types of information to feed their models especially emotional analysis of people. Emotions that people express on social networks, especially Twitter. The purpose of this paper is to introduce a new approach to Bitcoin trend prediction using deep learning algorithms. By sentiment analysis of extracted data from Twitter and tracking the previous price. The data collected for this research is between January 2012 and December 2020. This article compares LSTM, Bi-LSTM, GRU, and Bi-GRU algorithms to predict the trend of Bitcoin price changes. The Bi-GRU algorithm better performance by registering a record of 72% accuracy in predicting the trend of Bitcoin price changes and improving 20% the speed of the learning process.