Financial Data Prediction And Analysis Based On LSTM Neural Network

The financial market is constantly changing,and is affected by many factors suchas financial markets and political factors,which makes financial data also extremelyuncertain.However,predictive analysis and predictive modeling of financial data areof great significance.From a macro level,effectively predicting changes and trends infinancial data is conducive to national development strategies and economicdevelopment strategies by the country.From a meso level,Being able to understandthe changes in financial data is conducive to the planning of business strategies andgaining a competitive advantage in the market.From a micro level,individualinvestors will pay attention to historical data changes and the development ofindustries or companies before making investment decisions in order to obtainbenefits.The changes in financial data are always concerned by the public.Thefinancial data prediction problem is still an urgent problem that is closely related toindividuals and countries.Over the years,a large number of scholars have paidattention to the issue of quantitative investment,which has been accompanied by thecontinuous proposal and improvement of financial data prediction models.In recentyears,neural network models have been widely used in the prediction of financialmarket changes,and traditional time series models have also been enduring in thisregard,and are still widely used.This paper innovatively proposes a traditional timeseries model to improve the deep learning model.This article focuses on the improvement of the financial data prediction modelbased on the LSTM neural network.Taking Ping An Bank's stock price data as theexperimental object,it uses R language and python tools for experimental simulation.By analyzing the characteristics of financial data and the network structure of LSTM,it is found that financial data has the problems of high noise and nonlinearity,and theselection of LSTM parameters is affected by subjective consciousness.On the basis ofstudying the improved models proposed by past scholars,the ARIMA-PSO-LSTMmodel based on the ARIMA model and the PSO optimization algorithm is proposed.After the steps of stationarity test,model ordering,residual error test,model selection,etc.,the ARIMA model is established on the basis of the original data,and the residualerror of the dataset is output,which is used as the input of LSTM,and the PSOoptimization algorithm is introduced.The algorithm determines the parameters of datasegmentation and LSTM parameters.Complete the establishment of the PSO-LSTMcombination model,and then fit the stock price of Ping An Bank.The finalexperimental evaluation results show that,compared with the PSO-LSTM model andthe LSTM model,the ARIMA-PSO-LSTM model has smaller prediction errors and arelatively better fitting effect.The improvement of the model is effective.