A Hybrid Deep Neural Network Model for Stock Prediction

In this paper, we propose a hybrid deep neural network model (HDNNM) to predict the stock price. The HDNNM consists of two parts: classification and regression. The classification part utilizes convolutional neural network (CNN) to build a three - category classification model, which aims to predict whether the trend of stock in next day is up, flat or down. Then according to the result of the classification, data of different trends are fed into their corresponding regression models based on CNN-long short-term memory (CNN-LSTM), which aims to put the features extracted from CNN into LSTM for the price spread prediction. Finally, the predicted price spread plus close price of the day is the close price of next day. The proposed model was compared and verified by Standard & Poor's 500 (S&P 500) index, a stock index of 500 publicly traded companies in the United States. We evaluate the experiments by four technical indicators and the results show that the proposed method has better performance compared with other models.