Research On Stock Index Prediction Model Based On LR-LightGBM Model

The stock index is unpredictable,and the price fluctuation is affected by a large number of factors,generating massive amounts of data all the time.Therefore,processing tens of millions of indicators through machine learning algorithms as a reference basis for investors and bringing more benefits to investors has become a very worthy research issue.Based on the LightGBM algorithm,this thesis further optimizes the model and builds the LR-LightGBM model to predict the rise and fall of stock indexes,and the effect of the model has been effectively improved.First,the financial indicators are analyzed and compared,and 11 main technical indicators are selected as input features,which comprehensively reflect the various influencing factors that affect the changes in stock indexes.Select the CSI 500 Index and the Shanghai and Shenzhen 300 Index to predict and analyze the rise and fall of the stock indexes of two types of companies,which fully reflects the laws of large market value companies and small and medium market value companies.To verify the effect of the LR-LightGBM model in stock index prediction,the data is first processed through the LR model to obtain a composite evaluation index,and input into the LightGBM model together with other data for prediction.Different model structures have different preferences for different laws of the data.By processing the data through the linear model,the laws reflected by the linear model can be added to the tree structure model,which is conducive to deep mining of the data characteristics of the input model.,To achieve better results.Finally,compare the accuracy of the LR-LightGBM model with the LR,SVM,and random forest models,and predict the data of each year from 2016 to 2019.In different time dimensions,the LR-LightGBM model is compared with other models.The models have achieved good accuracy.Finally,back-test the LightGBM and LR-LightGBM models respectively.The results show that the effect of the LR-LightGBM model is better than the prediction effect of the original model,and it has been effectively improved in the backtest of actual stock data.