Research On The Forecast Model Of SSE 50 Index Based On LSTM

Securities investment is one of the most important investment and financial management option at the moment.The changes in stock prices are affected by many factors,on the other hand stock indexes can comprehensively measure and reflect the overall price level and changes in the stock market in a timely manner.Therefore,an effective stock index prediction can provide reference value for investors' investment decision-making and have important practical significance.In this paper,a prediction model named IPSO-LSTM for the SSE 50Index(namely Shanghai Stock Exchange 50 Index)based on the Long Short-Term Memory Neural Network(LSTM),is constructed.As a research hotspot in the field of deep learning in recent years,LSTM can process complex and long-term dependent nonlinear time series data,discover the hidden information in nonlinear data,and have good generalization capabilities.The specific methods are as follows:The future trend of the SSE 50 stock index is divided into two types: ups and downs,and the index's ups and downs in the next trading day are predicted through the historical data indicators of the trading day in the previous ten days.The following is given including four aspects: Firstly,in the selection and processing of indicator data,there are 5 transaction data indicators and 19 technical indicators,at the same time in order to unify the dimensions the maximum and minimum normalization method is used in the paper;secondly,in order to eliminate redundant information between multi-dimensional input data and avoid information to be overlapped,two feature extraction methods including principal component analysis(PCA)and autoencoder(AE)are used to reduce the dimensionality of the data in the paper.Experiments show that the input feature extracted by the autoencoder has better effect;thirdly,on the basis of the LSTM model,in order to improve the prediction accuracy of the model,particle swarm optimization(PSO)and a new particle swarm optimization(IPSO)are used to optimize the model parameters(such as the batch size,the number of hidden layer neurons,the learning rate and other key parameters)to determine the final LSTM Model structure;Finally,experiments show that classification accuracy rate of the IPSO-LSTM model based on autoencoder feature extraction,which reaches 66.10%,is higher than the stock index prediction model based on support vector machine.