Shanghai Composite Index Prediction Based On Baidu Index And Random Forest

The stock market is a nonlinear system with intricate variable relationships.The stock market trend prediction is a hot academic issue in cross-disciplinary research such as finance,statistics,and machine learning.How to accurately predict its future price or trend is a subject worth studying.The trend of the stock market is not entirely determined by its own internal laws,but also affected by the attention of investors.The thesis proposes a Shanghai Composite Index prediction method based on the Baidu Index and random forest.Based on Baidu index and random forest,a regression prediction model and a classification model of Shanghai Stock Index was proposed.The main work of the thesis is as follows:(1)The thesis using the Baidu index of the stock search keywords as an indicator of the investor's attention,and realizes the quantification of the investor's attention and its combination with the prediction of the trend of the Shanghai Composite Index.(2)Proposed a regression forecast model for the closing value of the Shanghai Composite Index based on Baidu index and random forest.The model screens stock search keywords through time difference correlation analysis,and combines the data of the Shanghai Stock Index attributes to propose the model.The thesis uses grid search and cross-validation to find the optimal parameters,and verify the model fitting effect by fitting coefficients.Compared with the traditional KNN,SVM classic regression model,this model significantly reduces the prediction error.At the same time,compared with the random forest regression prediction model without Baidu index,this model has higher accuracy and better fitting effect,which proves that the Baidu index is highly effective in predicting the model.(3)Proposed a prediction model of the rise and fall of Shanghai Composite Index based on Baidu index and random forest.On the basis of leading search keyword optimization,a feature selection method based on RF-RFE and cross-validation is proposed to select the optimal feature subset suitable for the model.The thesis uses grid search and cross-validation to find the optimal parameters.The OOB value,accuracy and F1 value are used to verify the prediction effect of the model under the optimal parameter settings.Through the comparison experiment without Baidu index,it is verified that the model with Baidu index has higher prediction accuracy and F1 value.At the same time,compared with MLP and GBDT model,the results also prove the model proposed in the thesis is better than other models.