Exploring the Performance of the CNN-LSTM Model in Stock Prediction

Abstract

Stock prediction in the financial market helps investors formulate investment strategies. However, the stock prediction has the characteristics of high dimension and high noise, and traditional linear models have difficulty meeting this demand. Therefore, it is necessary to use deep learning algorithms with stronger processing power for prediction. This paper combines the advantages of convolutional neural network (CNN) in feature extraction and the advantages of long short-term memory network (LSTM) in large-time scale relationship prediction to construct a CNN-LSTM hybrid network. In order to evaluate the performance of the model, this paper selects the closing price of the Shanghai Stock Exchange (SSE) 50 index in the past five years for experimental prediction. This hybrid model was compared with the ARIMA model and a single neural network, and it was found that the prediction results were more accurate, only lower than the LSTM model, but the interpretability of the prediction results was lower, only higher than the CNN model. This paper can provide ideas for the improvement of CNN-LSTM in the field of prediction. It also provides a reference for investors to decide on investment strategies and create a healthier stock market.