Research On Prediction Of Stock Price Index Based On Attention-BiGRU Model

As one of the hot spots in the financial field,the stock market constitutes a data system with a huge amount of data and intricate relationships between various factors,making stock price predictions expected to be further excavated and studied.Mathematical statistical methods are difficult to further deal with nonlinear relationships in practical applications,so it is difficult to explore the depth of stock information.At the same time,machine learning methods with outstanding application effects in other fields have been tried to be applied to the stock market,especially the application of neural network models and their composite models in stock prediction research is remarkable,and scholars have also found that the method has a poor grasp of important information about data in the process of applying neural network models to predict stock prices.In response to the above problems,scholars are exploring better neural network models while also combining other methods to jointly analyze stock data.Based on this,this paper uses the Attention-Bi GRU model constructed by combining the attention mechanism with the two-way gated looping unit network on the extraction of data features for the prediction of stock prices.The SARIMA time series forecasting method,LSTM model and variant model Bi LSTM,GRU model and variant Bi GRU model,and Attention-Bi LSTM model were used as control experiments.In order to make the research data convincing and representative,three different stock price indices of the CSI300 Index,the Shanghai Composite Index and the SZSE Composite Index were selected to predict the closing price on different models.The selection of important hyperparameters in neural network-related models is explained in detail.On the one hand,the research in this paper verifies the prediction performance of some neural network correlation methods and traditional statistical methods in stock price prediction,and on the other hand,explores the influence of attention mechanism on bidirectional gated circular unit model on stock price prediction.According to the empirical research,it is concluded that the Attention-Bi GRU model has a good fitting effect with low prediction error on three exponential stock prices,which improves the learning efficiency of the neural network model and shows good prediction stability,indicating that the Adaptation-Bi GRU model has a high adaptability to stock price prediction.This paper also gives relevant explanations and suggestions for improvement on the problems existing in empirical research.