Research On Stock Price Prediction Based On LSTM Model

The continuous increase in per capita income makes more residents choose stocks as a new investment method,so how to more accurately judge their price trends has become increasingly important,traditional time series analysis models such as ARIMA and GARCH have more restrictive conditions and can't adapt well to the growing stock market.With the vigorous development of neural networks,its powerful information processing capabilities have more and more applications in real life,different types of neural networks are suitable for analyzing different types of data,and the model suitable for time series data such as stock prices is Recurrent Neural Network(RNN).Although RNN can learn the context of the sequence,there is a long-term dependence problem,scholars used the Long Short-Term Memory(LSTM)model to improve it,the newly added Gating Mechanism in each unit of LSTM can make timely adjustments to long-term memory based on historical information and current input;in recent years the attention mechanism based on the "encoder-decoder" framework has gradually emerged and the perfect match with various neural network models makes it play an increasingly important role in many fields,we combine it with the LSTM model as an improved model to predict stock prices.In this paper,stocks are selected as the research object in the common application fields of artificial intelligence,a total of 35 stock price index variables are selected as the research features and added to the batch-normalization LSTM model,it's found that the prediction accuracy is not high;in order to reduce the complexity of the model,principal component analysis and deep autoencoder network are used to reduce the dimensionality of stock price features,the comparison shows that the information obtained by using deep autoencoder network is more concise,and improvs the prediction accuracy of the LSTM model;in view of the unstable prediction effect of this model,we decide to combine the attention mechanism with the LSTM model to improve it,the experimental results show that the improved model significantly improves the prediction accuracy,finally the stability analysis of the model was carried out on several other stocks,to live up to expectations,the decision cofficients predicted by the model are all over 0.9,and the stock price fitting curve is closer to the true trend.