Research On Stock Price Prediction Modeling Based On CNN-LSTM And Transfer Learning

China’s stock market is an extraordinarily significant part of China’s financial market.The accurate prediction of stock price is of great significance for investors to reduce the blindness of investment behavior,enterprises to maintain the financing function and countries to develop economy healthily and stably.In recent ten years,the rapid development of computer technology makes the deep learning model achieve fair results in dealing with complex nonlinear problems.Therefore,the combination model based on CNN-LSTM constructed in this paper is of certain research significance for stock price prediction.However,one of the main problems perplexing deep learning algorithm training is the problem of small amount of data.However,in the field of image processing,transfer learning has achieved a good effect in solving the problem of less data,so it is also of great research significance to apply transfer learning to stocks with less data.In this paper,the CNN-LSTM combination model is constructed by means of joint loss and applied to the stock price prediction of Shenzhen Component Index,CSI 300,Internet Finance and Artificial Intelligence 50.And through inductive transfer learning,the optimal CNN-LSTM combination model based on the Shenzhen Component Index and CSI 300 is applied to the Internet Finance and Artificial Intelligence 50 with less data,so as to obtain the high-precision stock price prediction model based on transfer learning.Firstly,the data of the highest price,lowest price,opening price,closing price and trading volume of four kinds of stocks are preprocessed to form stock image data fused with K-chart and trading volume bar chart and stock time series data in the form of logarithmic rate of return.Each piece of data contains 30 days of historical data,and the stock price index after5 days is predicted with a rolling window of 1 day.Secondly,the CNN model and LSTM model were constructed by setting the control group and changing the combination of internal hyperparameters in the model,and the prediction effect of the CNN-LSTM combination model under the combination of hyperparameters was compared and analyzed with the CNN model and the LSTM model.Finally,the optimal CNN-LSTM combination model based on Shenzhen Component Index and CSI 300 is used as the pre-training model for the second training of Internet Finance and AI 50,so as to obtain two kinds of high-precision stock price prediction models based on transfer learning of Internet Finance and AI 50.The prediction effect is compared with that of CNN model,LSTM model and CNN-LSTM combination model.The results of this paper show that:(1)the combined loss CNN-LSTM model constructed in this paper has a better performance in the stock price index prediction of the Shenzhen Component Index,the CSI 300,the Internet Finance and the Artificial Intelligence 50.Compared with the single neural network model,the prediction performance of this model based on the Shenzhen Component Index and CSI 300 is improved significantly.For the Internet Finance and Artificial Intelligence 50,which have a small amount of data,their performance is slightly improved compared with LSTM.(2)The predictive performance of the high-precision CNNLSTM combined model obtained by transfer learning in this paper is significantly improved compared with the combined model obtained by direct training and the single neural network model on the datasets of Internet Finance and Artificial Intelligence 50.