| Since Satoshi Nakamoto invented Bitcoin in 2009,social recognition and demand for Bitcoin and similar digital currencies have greatly increased with the development of Bitcoin and blockchain technology.The rapid rise in the price of Bitcoin and the characteristics of large fluctuations have attracted a large number of users to invest in it as a digital asset.Before the regulatory strategy of a standardized system has been formed,the development of Bitcoin will undoubtedly have an increasing impact on society,and the fluctuation of its price will become a social instability factor by increasing the risk of holding users.Therefore,finding out the factors that affect the price of Bitcoin and predicting its price have become a hot spot in Bitcoin research in recent years.This will not only help investors and related institutions understand the Bitcoin and digital currency markets,but also help improve the financial market and its policies.Based on the previous research on bitcoin and stock market price prediction,this dissertation uses a machine learning model to predict bitcoin prices in terms of value and trend.The main content includes the following aspects.First,in terms of data characteristics,a more comprehensive summary and reference to the previous domestic and foreign research on the price of Bitcoin,summarized into four aspects of bitcoin market information,blockchain information,macroeconomic information and search index information,and incorporate it into the model for analysis.Then,according to the huge difference between the price volatility of Bitcoin in the two time periods around 2017,the two time periods are divided into two time periods to conduct a longitudinal comparative study on the price of Bitcoin.Second,in terms of model prediction,a two-stage feature processing method is designed,using three Recurrent Neural Network models(RNN)and Convolutional Neural Network models(CNN)to compare and predict the price of Bitcoin,and using Recursive Features Elimination(RFE)and Logistic Regression(LR),Random Forest(RF),Linear Discriminant Analysis(LDA),Support Vector Machine(SVM)and Naive Bayes(NB)five commonly used machine learning models are combined to predict the Bitcoin price trend.Through empirical analysis,this dissertation obtains the following results.First of all,in terms of feature selection,this dissertation summarizes the influencing factors in the previous related research.The results of the model prediction in this dissertation show that the prediction accuracy is better than most previous researches.Second,in terms of Bitcoin price prediction,in order to reduce the impact of redundant input features on model prediction,this dissertation adopts a two-stage feature processing method of feature selection + dimensionality reduction,and finds the characteristics of factor analysis(FA)in two stages.The processing method has more advantages than the other two dimensionality reduction methods,and can effectively reduce the dimensionality of the related complex features.Third,as a typical time series,deep learning can better predict the price of Bitcoin,and RNN is better than CNN in predicting the price of Bitcoin.In the results,GRU and LSTM are better than CNN.Fourth,in terms of Bitcoin price trend prediction models,this dissertation uses a combination of Recursive Feature Elimination(RFE)and machine learning models.The prediction results show that Random Forest(RF)as an hybrid model is used in Bitcoin price trend prediction Shows better stability.Regarding the influencing factors of Bitcoin price,through the above empirical research,this dessertation draws the following two conclusions.First of all,there are many factors that affect the price of Bitcoin.Among them,the market transaction information and blockchain information of Bitcoin play a vital role in model prediction.Secondly,affected by the social acceptance of Bitcoin,the price of Bitcoin fluctuates greatly at different stages of development.In addition to Bitcoin and blockchain information,other major influencing factors are also different.Through the division of Bitcoin range and empirical research,it is found that before2017,oil price information has a significant impact on Bitcoin price trend prediction.After 2017,Baidu index information has a significant impact on Bitcoin price trend forecasting,which shows that after 2017,as the popularity of Bitcoin and blockchain in China increases,Chinese people are paying attention to Bitcoin that affects Bitcoin’s price changes.In general,this dissertation comprehensively considers the impact of Bitcoin prices based on previous research,and predicts Bitcoin prices and trends at different stages through deep learning and machine learning models,and discovers the main factors that affect Bitcoin prices.The research not only provides basis and ideas for Bitcoin price and related research,but also at the current stage of rapid development of the Bitcoin and digital currency market,it helps all walks of life understand the characteristics of this type of market,and provides a basis for the country to formulate relevant policies such as supervision. |
