Stock Price Prediction Using Kernel Principal Component Analysis And Support Vector Regression On Daily And Up To The Minute Prices

With the development of reform and opening-up,China's stock market is gradually recognized by the world's major economies,and the domestic stock market has attracted more and more foreign investors.The purpose of the stock price prediction system is to provide abnormal returns for financial market operators and become the basis of risk management tools.With the advent of the era of big data,computer information technology and financial engineering are increasingly connected.In the development of stock trading mechanisms,computer-intensive systems that use machine learning algorithms are becoming more and more common.Many scholars at home and abroad use statistical learning methods to establish stock price prediction models under the support of scientific theories,and they have been applied to stock market price and price trend prediction.Based on support vector regression(SVR),this paper introduces a nonlinear learning technique to combine dimensionality reduction methods into an integrated learning model to predict the Chinese stock market index.This topic has certain research significance and application prospects.In this paper,the stock index of the Shanghai Stock Exchange is taken as the subject of research,and the SVR model theory is applied to the prediction of prices at different frequencies in the stock market.First,based on the stock price prediction problem,daily data and minute data are selected for model parameter selection,a nonlinear support vector regression prediction model is constructed,and the converted closing price is predicted.Before training the model,according to domestic and foreign research literature,select appropriate TA technical indicators as model input variables.Considering that different combinations of data feature variables will show different prediction results during model prediction,in order to further prevent feature variables from falling into the trap of "high-dimensional curse",kernel principal component analysis of nonlinear technology is introduced.Then,the cross-validation method is used to optimize the combined parameters of the pre-processed data set based on the grid search method.The parameter optimization combination studied in this paper includes the number of PCs in the kernel principal component analysis,Gaussian kernel parameters and nonlinear SVR Penalty coefficient in the model,threshold of loss function,nonlinear kernel parameters.Finally,the optimal parameter combination is selected through the combined parameter selection theory,and the optimal parameters are used to model the stock data sets of different frequencies.It is expected that during the stock price prediction process,the stock will make a high value in the next trading period.Precise and low-error stock price prediction,which provides investors with more trading opportunities and profits.The stock price prediction problem involves complex time series and nonlinear problems,and it is easily affected by various unpredictable factors during the modeling process.This article takes stock prices of different frequencies as the research object,finds suitable technical indicators from the constantly changing stock market,and uses nonlinear feature dimensionality reduction techniques and SVR machine learning algorithms to focus on modeling and analysis of highfrequency data.The model parameter adjustment method to improve the accuracy of stock price prediction proves the feasibility of the research method proposed in this paper and provides some reference for investors' investment strategies.