Stock Price Trend Predicting Research Based On Machine Learning Method

The studies of stock price trend predicting have never stopped since the birth date of stock market.Predict stock price trend accurately may bring huge returns for investors under fewer investment risks.The stock market is a complex nonlinear system.Traditional analysis of financial time series has limitation in stock predicting.Machine learning methods have powerful ability for nonlinear problems.In recent years,machine learning methods are widely applied to financial time series predicting.It has become popular in the study of stock price trend predicting.Compared with traditional modeling methods,the methods based on machine learning algorithm have unique advantages in stock trend predicting.In this study,first,the feasibility and advantages of the machine learning method in the research of stock price trend predicting were discussed and pointed out the advantages and disadvantages of the existing learning algorithms used in stock predicting.The theories of statistical learning and support vector machine were introduced.Then,the theory of twin support vector machine was discussed and the stock trend predicting model based on TWSVM algorithm was proposed.Shanghai Securities Composite Index(SSCI)and Standard and Poor's 500 Index(S&P500 Index)daily trend were the targets of the predicting tasks.Predicting models based on other five machine learning algorithms are set as contrast experiments,including Decision tree(DT),Naive-Bays(NB),random forests(RF),probabilistic neural network(PNN)and support vector machine(SVM).The experiment results indicate that TWSVM predicting model has a better predicting performance on both stock price and index daily movement.Finally,the importance of feature selection was discussed.and a two step feature selection and stock trend predicting model(DFS-BPSO-SVM)was proposed.In this process,a criterion of DFS(discernibility of feature subsets)combined with feature subsets predicting performance was applied for feature selection.Compared with another 5 feature selection and predicting model PCA-SVM,Relief-SVM,GA-SVM DFS-BPSO,SVM respectively,the experiment results showed that the predicting moedel based on DFS-BPSO-SVM has a better performance on SSCI daily trend predicting.