Predicting Bitcoin Price Based On Deep Learning Method

Known as the "digital gold",Bitcoin is the first digital currency to be widely recognized around the world.Researchers’ early studies on Bitcoin mostly focused on it’s technology and principle,but with the accumulation of data related to the price of Bitcoin,people gradually began to analyze and predict the price trend of Bitcoin from the perspective of machine learning,so as to guide investment decisions.With the maturity of deep learning technology,many deep learning models have been applied to the prediction of time series data and achieved good results.But there are still a number of problems in predicting the price of Bitcoin.It is mainly about how to reduce the time-lagging problem effectively,how to use external information to improve the accuracy of prediction effectively,and how to establish an efficient prediction model.In this paper,a comparative experiment is conducted between the traditional machine learning model HMM and the deep learning model LSTM to demonstrate the powerful power of deep learning model in the field of time series data prediction.In the experiment of introducing the traditional financial market data such as Shanghai Stock Index,Dow Jones Index and gold price into LSTM,it is found that the gold price has a certain correlation with the trend of bitcoin price.In the experiment of using BERT to analyze the public sentiment data of Bitcoin forum and adding it as a feature into the LSTM model,it is found that the effective use of public sentiment data can improve the prediction effect of the model.This paper designed a two-layer model of LSTM+GRU based on Stacking idea and rolling training.By setting up different training methods,the model obtained different perspectives,effectively utilized the historical information,and further improved the prediction effect of the model.