Digital Currency Price Prediction Based On CEEMDAN＿SE＿PSO＿ELM Composite Model

This thesis first proposes the development of digital currency based on the foundation of blockchain technology and the background of the digital economy,which leads to the significance of digital currency price prediction.Focusing on the problems related to digital currency price prediction,whether it can be measured,what method is used to measure,and the measurement is accurate or not,a brief literature review is made in these three aspects.Next,this thesis first makes descriptive statistics on the basic situation of digital currency,and explores the factors affecting the price of digital currency to understand the overall trend of its price and the law of volatility in the time series.After that,this thesis uses the CEEMDAN-SE-PSO-ELM composite model to predict the price of multiple digital currencies.It not only predicts the most common bitcoin price,but also conducts certain research and analysis on other digital currencies.The results discover the overall law of digital currency price prediction and improve the prediction accuracy of current digital currency machine learning algorithms,which have a wide range of application significance.This thesis analyzes and predicts the prices of the top five digital currencies(Bitcoin,Ethereum,Cardano,Binance Coin,TEDA)in the digital currency market,with the daily closing prices from July 1,2017 to June 30,2021 as the original sample data.The CEEMDAN-SE-PSO-ELM algorithm is used to calculate the corresponding average absolute error MAE value.The result shows that the MAE value of the composite algorithm proposed in this paper is much smaller than the average daily price change of the sample data,which proves that our composite model is effective and accurate.Finally,this thesis uses other common machine learning algorithms to predict the same sample data,and integrates the MAE results of all models for comparison.It turns out that the MAE value of the model in this thesis is smaller than other models,which proves that the CEEMDAN-SE-PSO-ELM composite model proposed in this thesis for the price prediction of digital currencies can greatly improve the prediction accuracy.Among them,the optimization algorithm has a greater contribution,and the decomposition algorithm can also improve the accuracy by a small amount.At the end of this thesis,we also make research prospects from the three aspects of replacing and improving machine learning algorithms;making more adjustments to evaluation indicators;and studying more indicators of digital currency besides price,so as to provide references for future research.