Research On The Application Of Data Mining In Stock Selection

After nearly thirty years of continuous development in China's stock market,the stock market is gradually becoming mature,and the analysis methods and analysis methods of stock market tend to be diversified.The traditional analysis methods of stock include the basic analysis method and the technical surface analysis method.With the continuous expansion of stock market data,traditional processing methods consume much time when dealing with large amounts of data,and more often rely on human judgment,which is difficult to quickly discover the drawbacks of the regularity between data.With the continuous development of AI,data mining and machine learning technology,the application of such technologies to stock data analysis is constantly emerging at home and abroad.The use of these new methods of data mining is to obtain stocks that exceed stock benchmarks.In this paper,four kinds of data mining techniques are selected from the multi data mining technology to find the stock characteristics with excess income.The data mining methods used include association rules,principal component analysis,decision tree model and deep neural network model.The characteristics of the selected stocks include five indicators,which are financial reporting indicators reflecting the company's overall financial status,external industry indicators,technical indicators,consistent expectations class indicators and valuation indicators.Because all kinds of indicators have strong correlation inside,if all indexes are input into data mining model at the same time,it is easy to cause the problem of parameter estimation caused by multiple collinearity.So this paper will use the methods of association rules and principal component analysis of two kinds of dimensionality reduction of data preprocess data,through the high concentration of dimensionality reduction as input variables are input into the decision tree model and the depth of the neural network model,to forecast the stock excess returns relative to the CSI 300 refers to the number of.Through data validation,this paper finds that the accuracy of data prediction and ROC curves are better than those based on association rules.The prediction model has significant excess returns,and the prediction effect through the deep neural network model is better than the prediction effect based on the decision tree model.From the forecast period,the excess return of three month is significantly better than that of one week's excess return and six month's excess return.Therefore,the conclusion is: investors should say that the adjustment period is set to three month or so to achieve the best investment results.