Research On Optimization Of Stock Index Prediction Model Based On Random Forest And LSTM Network

In the financial market,the prediction of stock price index has always been a hot field.Due to the complexity and chaos of the price series of stock price index,it is difficult to fit the statistical measurement model which focuses on predicting causality.In addition,the signal-tonoise ratio of stock price index time series data is too low,and the direct use of stock price index market data prediction is often ineffective.Therefore,this paper uses machine learning and neural network model to predict stock price index.In the construction of data set,the market data,technical indicators and related Baidu index search keyword data that affect the fluctuation of stock price index are selected.The main research is as follows:(1)Research on stock index prediction model based on optimized random forest.For multi-dimensional stock price index time series data,the direct use of random forest algorithm cannot effectively screen a large number of features.Different features may have the same effect on the model,and even reduce the accuracy of the model.Invalid features will increase the training difficulty of the model.In this paper,a hierarchical feature selection method combined with P value check is designed for feature processing(Hierarchical Feature Selection Method Combined with P-value Verification,PFS),The features are divided into strong correlation feature set and non-strong correlation feature set.The features with low redundancy and strong cross-correlation are selected from the non-strong correlation feature set.The filtered composite feature set is verified by P value to obtain the final optimal feature collection.Input into Random Forest(Random Forest,RF)algorithm,build PFS-RF stock price index prediction model.By comparing the prediction effect of the model on different years of data sets and comparing the prediction effect of the model with the random forest model under different feature processing methods on the same data set,it is verified that the feature processing algorithm in this paper can effectively screen features and improve the prediction accuracy of the model.(2)Research on stock index prediction model based on optimized LSTM network.In order to shorten the training time of multi-dimensional stock price index time series data and reduce the training difficulty of LSTM network,this paper uses the calculation of feature importance in the training process of random forest algorithm to select features with large importance value.The premise that LSTM network is suitable for stock index prediction is to select appropriate regression parameters,use Whale Optimization Algorithm(WOA)to optimize the important parameters of LSTM network,and set the other parameters through comparative experiment and experience.Finally,the RF-WOA-LSTM stock index prediction model is proposed.By comparing the prediction effect of the model on data sets of different years and the prediction effect of the model and the commonly used stock index regression model on the same data set,the validity of the model is verified.