A EEMD-PCA-SVR Model And Its Application In Stock Price Prediction

Stocks attract the attention of different groups of people because of their high yield and high risk.According to the "National Stock Market Investor Status Survey Report",the number of stock investors in the past three years was 14,650.44,15,975.24,and 17,777.49,an annual increase of nearly 9%.In addition,the impact of stocks on the country’s economic and social life is increasing.It can not only fully gather idle funds in the society,but also circulate funds in different regions of the country,and can also absorb foreign capital.No matter from the interests of shareholders or social and economic development,stock price forecasting is particularly important.Aiming at the nonlinear,non-stationary and high-complexity time series characteristics of stocks,this paper proposes an EEMD-PCA-SVR model for stock price forecasting by combining the Ensemble Empirical Mode Decomposition(EEMD)and Support Vector Regression(SVR)models.By using the difference of the original stock sequence data to replace the input,the change trend of the stock price is highlighted,and the sequence after the difference is processed by the ensemble empirical mode decomposition to obtain a number of low-complexity and simple-structured eigenmode functions and other items.;Principal component analysis is used to reconstruct the eigenmode function to reduce the workload of prediction,and obtain high-frequency mode,low-frequency mode and remainder;Support vector regression model predicts high-frequency mode and low-frequency mode,ARIMA model predicts the remainder,and sums There are three types of forecast results,and the stock price forecast value is obtained.The data of the closing prices of the Southern China Securities 500 Composite Index stocks from March 9,2015 to November 29,2019 were selected for the experiment.The differentiated Southern CSI 500 Composite Index was decomposed into eight eigenmode functions(IMFs)and one remainder by EEMD technology;the first three of the eight IMFs were combined according to the cumulative variance contribution rate by principal component analysis.It becomes the high-frequency modal HIMF,and the remaining 5 IMFs are added together to form the low-frequency modal LIMF;when using the support vector regression machine to predict,use grid search and cross-validation to find the three parameters in the support vector regression model.The optimal parameters of the high-frequency and lowfrequency support vector regression models are obtained.The prediction results show that the model proposed in this paper has a score of 0.00049,0.0189,and 0.6274 in root mean square error,mean absolute percent error,and direction statistics,respectively,which are better than the single SVR model,the IMF-based model and the mean reconstruction model.It can provide a certain reference value for stock price prediction and has certain practical significance.