Application Of Fusion Learning Model In Security Investment

The security price prediction is one of the most important topics in financial statistics.Accurate security price prediction is of great value to market managers,investment institutions,listed companies and investors.Due to the complexity of the causes of financial market fluctuations,the previous single machine learning model has encountered the bottleneck of further development in the research process,and the fusion of different methods has brought new hope for the improvement of prediction accuracy.Therefore,this thesis attempts to use extreme gradient boosting(XGBoost)algorithm and support vector regression(SVR)algorithm to predict the security price.This paper is divided into two steps to predict the securities price.The first step is to establish the prediction model of the direction of the stock price based on the technical analysis,that is to select the stock characteristics,such as the closing price,the highest price and other trading data to build the corresponding technical indicators,and then use XGBoost to predict the direction of the securities price,including the direction prediction of the next 3 days,15 days,30 days,60 days,90 days and 120 days;The prediction results show that with the increase of time,the accuracy of prediction direction is also improved;According to the prediction labels of training set and test set,the data are divided into “1 label training data”,“0 label training data”,“1 label test data” and“0 label test data”.In the second step,according to the “1-label training data” and “0-label training data” obtained in the first step,the SVR model is used to predict the stock price respectively,and then the “1-label test data” and “0-label test data” are brought into the model with the same corresponding label to obtain the securities price of the test set in different time periods.In addition,grid search is used to optimize the parameters in the training model.Finally,the corresponding trend investment strategy is constructed by using the prediction results,and the simulation investment is carried out,and good returns are obtained.By comparing the experimental results of different length of time,it is found that long-term investment is more likely to bring greater benefits in trend investment.In order to further illustrate the effectiveness of the model,this thesis compares the prediction result of XGBoost-SVR model with that of the deep stock trend prediction neural network model(dspnn-mt)trained by weighted market information and comprehensive trading information of multiple stocks,the accuracy of direction prediction is significantly improved.