Stock Short Term Prediction Based On LSTM Neural Networks

Stock price prediction is considered as one of the most challenging problems in time series forecasting.It has important research significance and application value for the analysis and prediction of the future change trend of the stock market.Artificial neural network(ANN)algorithm is widely used in the stock market because of its strong selflearning ability,and better performance than traditional statistical and econometric models.In recent years,deep learning based on artificial neural networks has developed rapidly,and achieved unprecedented success in the fields of voice and image,but it is rarely used in the stock market.At present,the deep learning algorithm represented by autoencoder has made remarkable achievements in forecasting problems.The autoencoder generally adopts the two stage learning framework: 1)unsupervised feature learning process;2)supervised prediction learning.This framework can not only g take the advantages of deep learning in feature learning,but also achieve good performance in prediction.The paper proposes a deep learning framework represented by autoencoder to predict short-term stock prices.The main research work is as follows:1.In order to fully considering the factors related to the stock price,the paper selects Close price,Open price,High price,Low price and the technical index,and sets up the index system for the short-term forecast of the stock price.2.In the stage of unsupervised feature learning,because stock data has the characteristic of high dimension and noise.In order to reduce the noise and dimension,the paper combines the characteristics of the automatic encoder to improve the autoencoder network structure.In order to learn the depth feature of the stock price,the paper builds an unsupervised feature learning model of the stacked autoencoder(SCAEs)based on the improved autoencoder.3.In the supervised prediction stage,to overcome the shortcoming of Recurrent Neural Network(RNN),the paper introduce the Long-Short term memory neural network(LSTM)to improve the accuracy of stock price prediction.Finally the paper proposes the SCAEs-LSTM algorithm.The paper analyzes the impact of SCAEs-LSTM's parameters on stock price prediction on the S&P 500 index,and selects the appropriate parameters.In order to measure the performance of the SCAEs-LSTM algorithm,the paper selects three data sets of the S&P 500 index,the Hang Seng Index and the Shanghai stock index to compare the prediction effects of SCAEs-LSTM,LSTM and RNN on the three data sets,using python and tensorflow to accomplish the algorithms.The experimental results show that the SCAEs-LSTM algorithm can achieve the best prediction performance,and the more mature the stock market,the better the prediction effect.