The Appliance Of ANN And SVM Combining Relief Algorithm In Stock Price Index Forecasting

As an important part of the securities market,stock price changes affect many par?ticipants in the stock market.If we can predict the direction of stock price successfully,it is of great significance for stock market.For speculators,we can predict the direction of stock price and gain short-term profits in many transactions.For investors,we can better judge the market situation and make corresponding investment strategies by pre-dicting the direction of stock price.For risk management,it is more advantageous to predict the direction of stock price from the unreasonable stock price movement,and we can find market risks and make corresponding control and adjustment.Therefore,finding effective methods to predict the price of stocks or fluctuations has become a focus of attention in the financial sector.However,the prediction of stock price is a very complicated problem in the finan-cial field,because the movement of stock price is influenced by many factors,such as political events,company policies,general economic environment,investment expec-tations,the choice of institutional investors and investment psychology.These reasons lead to the trend of stock price movement is a non-linear,complex and chaotic system.Therefore,we need complex models to solve the problem of stock price prediction.Although stock price forecasting is very difficult,it has not stopped people's enthusiasm for research on stock price forecasting.The methods used so far are generally divided into basic methods,traditional statistical methods,machine learning methods and other methods.This paper will start with the application of BP Neural Network in price forecasting of CSI 300 Index,and discuss whether different machine learning models,bigger data size and Relief algorithm are conducive to improving the accuracy of model forecasting.In order to achieve the above purpose,the main work of this paper is divided into three parts.In the first part,BP,RBF neural network and SVM(Support Vector Machine)are applied to daily frequency data of CSI 300 Index to compare the prediction ability of different models.In the second part,BP neural network is applied to daily frequency data and minute frequency data of CSI 300 respectively to verify whether data size can improve the prediction ability of BP.In the third part,Relief algorithm is combined with BP,RBF neural network and SVM in the daily frequency data of CSI 300 index to verify whether Relief feature selection can improve the prediction ability of the model.The first part of the empirical results shows that the prediction ability of SVM is higher than that of RBF neural network,and RBF is higher than that of BP neural net-work.The second part shows that the prediction ability of BP neural network will be improved with the increase of data volume.The third part shows that Relief algorithm can improve the prediction ability of SVM(polynomial kernel function).