Prediction And Research Of Market Index Based On Machine Learning

Stock index is a nonlinear dynamic time series,which has the characteristics of high fluctuation,low stable noise and easy to be interfered by external factors.These characteristics make the accurate prediction of stock index become a very challenging problem.It is found that there are some problems such as multicollinearity and noise interference among the basic data characteristics of stock index,which often lead to a serious decrease in the accuracy of stock index prediction model;In addition,the research also shows that there is a large difference in the performance of machine learning models of different structures,which lead to different models of different models in the same stock index prediction.In this paper,the feature generation method based on Xgboost model and the dynamic weighted integrated learning model are proposed for the two factors that have the greatest influence on the stock index prediction.The main research content of this paper is as follows:(1)Research on feature generation method based on Xgboost model.It is found that the input characteristics have great influence on the performance of stock index prediction model.The existing methods of feature selection and feature extraction are partially missing and inadequate in the use of basic data information.In this study,it was found that when the Xgboost model projected the basic data characteristics of stock index into the high-dimensional space represented by leaf nodes,whether leaf nodes participated in the expression had an important impact on the prediction performance.In this paper,leaf node information expressed in Xgboot is extracted and one-hot coding is conducted to improve the expression ability of this feature.Experimental results show that the combination features generated by this method can effectively improve the accuracy of stock index regression prediction.(2)Research on stock index regression prediction method based on dynamic weighted ensemble learning.It is found that the prediction model also has significant influence on the prediction of stock index.In the traditional integrated learning model,there are some problems in stock index regression prediction,such as ignoring the contribution of the performance of the basic learner,and using the high performance basic classifier is limited.Because of the different structure of basic classifiers,their performance in different stock index forecast is different.In this paper,it is found that there is a certain degree of complementarity among the basic classifiers.By combining the basic classifiers with different structuresthrough dynamic weighting,the complementarity can be reasonably used to improve the contribution of high-performance classifiers.Based on this research,this paper presents a dynamic weighted integrated learning model for stock index prediction.The experimental results show that the dynamic weighted ensemble learning model proposed in this paper is more accurate than the single prediction model and is suitable for the regression prediction of different stock indexes.