Stock investment is one of the most popular means of investment in modern society.With the high expected return,its high risk can not be ignored.It is the practical need for investors to seek the best investment idea to avoid the bad.Investors and researchers have been trying to explore the changing trend and rules of stock price,hoping to predict the future stock price relatively accurately.From the traditional data statistical analysis methods to the existing deep neural network models,researchers are used to study the changes of stock price.At present,the extended research based on Long Short-Term Memory(LSTM)is a better research method for financial time series problem.Based on the LSTM model,it is a in-depth research for the stock price prediction problem.In order to solve the problems of over-fitting,gradient disappearance and model collapse in previous experimental studies,an improved prediction model based on LSTM-Adaboost was established in this paper.On the selection of model structure,the Dropout mechanism and introduce L2 regularization term are used;on the selection of activation function,PRe LU(Parametric Rectified Linear Unit)function is used;These measures can improve the prediction effect of the model;and enhance applicability of the model;The price data of two stocks and four main predictors of OHLC-Avg,RSI,MTM and MA were selected,so that our research is more representative and the prediction effect is better.Before the numerical experiment,all the input data are normalized and smoothed in advance,which can improve the robustness of the model.During the experiment,20 years of stock price historical data were input and divided into 48 pieces for cross-verification.In each piece,4 years of stock data were used to predict the stock closing price of 4 months,which could expand the experimental data volume,better adjust the parameters of the model,and make the prediction model more accurately.In order to measure the improved LSTM-Adaboost model established in this paper,the prediction effect,this article selects the root error RMSE,the average relative error MRE,R square shooting DS as evaluation index,and the Support Vector Regerssion(SVR),Back Propagation(BP)Neural Network,Recurrent Neural Network(RNN)and Long Short-Term Memory Neural Network,four stock price prediction models were compared.The results show that the improved LSTM-Adaboost network prediction model is better than the other four comparison models in stock price prediction.The prediction hit ratio of the model with better prediction performance in historical studies was about 59%,while the hit rate when using the LSTM-Adaboost network model for prediction is 64%,which is 5 percentage points higher than that of the others.Compared with the existing prediction models,the improved model hit ratio is significantly improved.The research work of this paper,in the theoretical aspect,has certain reference value to the future research of stock price prediction;In the practical aspect,it can also provide the decision-making reference for investors when they make stock investment. |