Prediction And Analysis Of SSE 50 Index Based On Support Vector Machine

China's stock market is inextricably linked to the national economy,and the stock market is a complex and nonlinear dynamic system.Stock data also has its own characteristics of non-stationarity,chaos,etc.The use of traditional time series analysis methods for modeling has great limitations,and machine learning methods have strong application capabilities when dealing with nonlinear problems.Support vector machine is a widely used technical method in machine learning.It presents its unique advantages in solving non-linearities,function fitting,high-dimensional pattern recognition,and prediction of time series.But how to improve the performance of support vector machines is still worthy of our in-depth study.Feature extraction is the first key to developing support vector machines,and the selection and optimization of kernel parameters is another key.This paper attempts to use kernel principal component analysis for feature extraction,genetic algorithm for parameter selection and optimization,and a comprehensive model of support vector machines to improve the predictive power of stock prices.The main research work of this paper can be divided into the following three parts:In the first part,the data of the SSE 50 Index was collected and characteristic indicators were determined,including 23 characteristic dimensions such as basic indicators and technical indicators,and relevant statistical analysis was performed on the determined data set.The results show that,except for a small number of characteristic indicators,such as random D%,the distribution of other characteristic indicators basically conforms to the normal distribution.Therefore,the distribution of financial data has no obvious law,and it cannot be predicted and analyzed by general linear regression methods.It needs to use machine learning to train and learn its internal laws.In the second part,the data of the SSE 50 Index from October 9,2009 to December 28,2018 were regression predicted and analyzed.In the implementation process,a combination model based on kernel principal component analysis and support vector machine using genetic algorithm for parameter optimization was used for regression prediction and analysis,and compared with the combination model based on principal component analysis and support vector machine,support vector machine model.The empirical results show that the performance of the comprehensive prediction model using KPCA and SVM is significantly better than the combined model using PCA and SVM,and SVM model,which verifies the effectiveness of the method.In the third part,we use the empirical results to obtain three models with certain parameters,and then perform regression prediction and analysis on the data of the Shanghai 50 Index from February 12,2019 to October 8,2019.The results show that the comprehensive prediction model of KPCA and SVM using genetic algorithm for parameter optimization not only has better prediction effect on the inside of the sample data,but also has better prediction effect on the related data outside the sample,which shows that it has a certain promotion ability.