Research On Stock Selection Model Based On Linear Discriminant Analysis And Weighted Support Vector Machine

Quantitative stock selection is an important branch of quantitative investment,the study is the selection of quality stocks.Since its birth has experienced more than 30 years of development,and in the Western stock market has achieved great success.With the continuous improvement and development of China's stock market,quantitative investment will also become the mainstream investment philosophy of China's stock market.In this context,this paper based on the classical theory of quantitative investment,the use of financial data of listed companies to establish support vector machine stock selection model for stock selection issues.In this paper,based on the study of classical western theories and multi-factor stock selection models,the paper analyzes and explores the selection of effective factors and the confirmation of weights,the establishment and optimization of models respectively.Finally,combining with ReliefF algorithm,Linear Discriminant Analysis(LDA)and Feature Weighted Support Vector Machine(FWSVM).First of all,ReliefF algorithm can calculate the weight vector of 54 financial indexes of seven categories by comparing Euclidean distance of the similarity of financial feature and nearest neighbor sample of different classes,and introducing Pearson correlation coefficient to eliminate redundant financial indicators,ReliefF algorithm cannot deal with the problem of feature redundancy,and obtain a scientific candidate feature set.Secondly,the linear discriminant analysis method is introduced,and the seven categories of financial indicators weighted by this paper are respectively projected to the best separation direction using the known class information.Finally,we establish the ReliefF-LDA-FWSVM classification model,and set up a classification model for different categories of samples.Then,we construct the ReliefF-LDA-FWSVM classification model for different categories of samples Quantity imbalance problem,the use of cost-sensitive principle to adjust the penalty parameters,to a certain extent,eliminate the impact of data imbalances,thereby enhancing the high-yield stocks recognition rate.In addition,the genetic algorithm is used to optimize the parameters of SVM,and the model is reasonable.In this paper,four years' rolling forecasting models are constructed using the structured data of Shanghai A shares Market from 2010 to 2014.Each model uses the financial data of the current year to identify the high-yielding stocks in the following year,and the comparative experiment is established.The experimental results show that the ReliefF-LDA-FWSVM model can effectively identify high-yield stocks.