Research On Stock Forecasting Model Based On Convolution Neural Networks

Financial market is a huge dynamic field,which is difficult to model and fore-cast.In recent decades,computational intelligence models that can learn the nonlinear relationship have been gradually used.Because it is difficult for tra-ditional methods to capture the nonlinear relationship between high-dimensional indicators and forecasting variables,deep learning model has more advantages in stock forecasting.In this paper,we use convolution neural network(CNN)in deep learning to build a stock forecasting model.On the basis of this model,the rise and fall of the stock are predicted.Based on the 15 day stock data,the rise and fall of the stock in the next week are predicted,and a good prediction effect is obtained.Firstly,the basic structure and back-propagation algorithm of convolutional neural network are explored.By introducing optimizer optimization training algorithm in parameter learning,and comparing several common CNN models horizontally,the convolutional neural network model is constructed on the basis of lenet-5 model.Secondly,this paper constructs a CNN based stock prediction model and makes an empirical analysis.The index data of 300 constituent stocks of Shanghai and Shenzhen from January 1,2010 to January 1,2020 are obtained from the wind database,and 15 different characteristics are selected to transform the sample data into a.two-dimensional matrix.The stock returns in the next week are divided into three categories:rising,fluctuating and falling,Innovatively use the sliding window method to increase the sample size,and divide all samples into five training sets and test sets.Then,the convolution neural network model is used to train and test all the stock data,and according to the convolution neural network structure,the model structure with more accurate prediction results is selected to optimize the parameters.At the same time,this paper considers adding optimizer to improve the back propagation algorithm to make the model more efficient.In addition,considering the imbalance of sample classification,this paper uses Rus technology to resample the sample data to improve the prediction ability of the model at the sample level.Finally,this paper evaluates the prediction results of the stock prediction model based on convolution neural network,and uses the accuracy rate and the average F-? to measure the prediction efficiency.At the same time,compared with the prediction results of LR model,SVM model and LSTM neural network,the results show that the average accuracy rate of the model constructed in this paper is better than the other three models,and from the performance of each year,The performance of this model is better than that of LR model.Therefore,the convolution neural network constructed in this paper is an effective stock forecasting model.