The Application Of Statistical Learning Methods In Financial Data Analysis

As the financial econometrics and financial statistics become more and more important,with the rapid updating of the global computer Internet and the large data processing technology,the financial data processing methods are gradually transferred from the traditional statistical theory to the field of artificial intelligence.A large number of financial information processing has also brought the source of the technological change,which has accelerated the in-depth study of the two levels of theory and application.Therefore,the research results of statistical learning methods in all aspects of the industry are becoming more and more fruitful,especially the extensive application of the theory of SVM,which provides a new solution for the data mining and the complex problems in economics.1.Using the grid search method,the stock price regression forecasting model of support vector machine is optimized.This model is mainly processed and analyzed by 2676 data in Shanghai and Shenzhen 300 from January 4,2007 to December 29,2017.Considering that the kernel function of SVM has a great influence on the accuracy of the model prediction results,the focus of this study is to optimize the parameter selection of the kernel function of support vector machine by using the grid search optimization algorithm.By comparing and analyzing,the optimized regression pretest model of the Shanghai and Shenzhen index is finally obtained.It is confirmed by experiment results that support vector regression is applied to predict the index and achieved good results.2.According to the classification of the factors of non financial listed companies of A shares,and the evaluation of the classification and benefit level of this kind of company is more reasonable and accurate,the content of this part is mainly based on the nuclear principal component and the sparse least squares SVM to establish a classification model.Because the least squares SVM has fast algorithm features in practical application,this paper also increases the difficulty of calculating the complexity of the complex problem by increasing the sparsity of the least squares SVM from the algorithm,and also improves the accuracy of the analysis processing.This study established a multi classification model through KPCA and SLS-SVM,and divided it into four categories according to the characteristics of the existing listed companies.The experimental results show that the prediction accuracy of this classification model can reach 82%.