Study On Stock Index Prediction Based On Empirical Mode Decomposition And CNN-LSTM Neural Network Hybrid Model

The stock market is a highly volatile,non-smooth and non-linear dynamic market.Stock price fluctuations are characterized by uncertainty,high returns,and high risks,making it an object of interest and pursuit for many investors.At present,one of the research hotspots in China’s securities market is how to improve stock price trend prediction performance.The complex and volatile stock market has raised investors’ requirements for market analysis capabilities.It is not only about applying traditional financial theory to the expanded market,but also about combining it with cutting-edge technology.Currently,the most common ones are fundamental analysis,technical analysis and stock forecasting based on theoretical knowledge related to statistics and artificial intelligence.In contrast,analytical methods based on deep learning and machine learning are better able to deal with non-smooth and non-linear problems and obtain better results in financial time series forecasting.Traditional stock price time series data contains both linear and nonlinear aspects,thus,traditional models have limitations in handling its nonlinear part,while deep learning complements the traditional models.In this thesis,the performance and accuracy of stock index prediction are studied and analyzed using deep learning and machine learning methods with stock index price as the prediction target.Based on this,the following research work is accomplished in this thesis.First,introduces Empirical Mode Decomposition(EMD)based on the characteristic nonlinear correlation of stock index forecasts.Empirical mode decomposition is to decompose the series according to certain laws and divide them into different time scales and trend scales,which effectively reduces the influence of complex data,plays a stabilizing role,and maximizes the use of information in the original data,further enhancing the model’s ability to learn longterm dependencies between data.In addition,EMD has an adaptive feature that does not depend on specific basis functions,making it more suitable for the analysis of some non-stationary and non-linear complex series.Second,a hybrid neural network model that combines convolutional neural network(CNN)and long and short term memory(LSTM)neural network is proposed based on the time-series nature of stock index prediction.Since the exponential data has strong nonlinear and high noise characteristics,CNN can use convolutional method and pooling method to obtain the basic characteristics of the data,reduce the scale and complexity of the original data,and reduce the amount of computation.In the choice of model structure,Dropout is used to prevent over-fitting.In the choice of activation function,Rectified Linear Unit(Re LU)function is chosen to improve the prediction effect and enhance the applicability of the model.LSTM is a modified model of RNN,which adopts the idea of temporal and directional processing and can handle the backward and forward relationship of the input data well,and LSTM neural network has long time memory due to its special gate structure.In this thesis,the LSTM technique is applied to analyze the characteristics of the exponential,and the LSTM network model is used to make predictions.The LSTM model hyperparameters are set,and the final LSTM model structure is analyzed in terms of time step,number of hidden layer neurons,learning rate,etc.to obtain the optimal stock index price prediction model,thus improving the performance of the prediction model.Third,takes the SSE index as an example.Before the experiment,all the input data are normalized and smoothed to improve the robustness of the model.The corresponding data are obtained through a sliding window and brought into the corresponding algorithm for the experiment.Root mean square error,mean absolute error,and mean absolute percentage error were used to evaluate the performance of the model.The prediction results of LSTM model and CNN-LSTM model for SSE index were compared,and the LSTM model,CNN-LSTM model and the model in this thesis were applied to the prediction of CSI 300 index respectively to test the generality of the model.Finally,through example analysis,it is proved that the established model can predict stock prices better and make better predictions for the CSI 300 index.The results show that the model has strong feature extraction ability,strong data processing ability and strong generalization,which can provide a better reference for stock price prediction.The research in this thesis can not only improve the accuracy of stock index price prediction and provide reliable and efficient trading information for long-term investment users,but also provide theoretical basis and decision support for institutional investors and other investors’ stock index price analysis.