Application Of Deep Learning In Stock Index Prediction

As the traditional time series model assumes many conditions and it can find out the linear relationship between the stock price and its relating factors but can't reflect the stock's non-linear relationship well,traditional one-factor and multi-factor recurrent neural networks have lower prediction accuracy.In order to better extract the data's non-linear relationship and improve the accuracy of stock price prediction,this paper introduced the attention mechanism into the neural network based on the structure of encoding-decoding neural network and proposed the structure of the recurrent neural network based on the attention mechanism.The stock index price forecasting model extracts the deep features of the factors that affect the price of the stock index.Compared with the one-factor recurrent network and the multi-factor recurrent network,the model improves the accuracy of the stock index price forecast.In order to speed up the training of the model and reduce the amount of model parameters,a neural network that removes decoding-layer was used in the specific experiments.Finally,the experimental results show that in the prediction of the per-minute price of the SZ50 Index and the HS300 Index,the accuracy of the network model is higher than that of other traditional network models,and the processingspeed of the data is accelerated,moreover,the prediction is very close to the network that does not remove the decoding-layer.The model was proved to be an effective method in prediction of stock index price.