The Application Of Improved LSTM Neural Network In Stock Forecast

Since the birth of the stock market,investors and practitioners have been devoted to the study of stock forecasting models.However,due to the complexity of the factors affecting stock price changes,the traditional prediction method is not effective in practical application.Until the emergence of neural network,its complex model structure makes it better than traditional prediction methods,and has better fitting ability to data.In the process of using neural network to train data,the optimization algorithm of gradient descent method is a very important one.In many algorithms,Adam algorithm is the most widely used algorithm.In practical applications,there is a shortage of weight loss in the Adam algorithm,which makes the result of prediction,the problem of over fitting,and make the final prediction result unsatisfactory.Aiming at the shortcomings of Adam algorithm,In this paper,we introduce the weight decay process to improve the Adam algorithm.This can prevent data from overfitting to a certain extent.The convergence analysis of the improved Adam algorithm.Through analysis,we can find that the difference between the cost sequence ft(θt)and the optimal cost is less than the constant number of the series(?).In the experiment,the LSTM neural network is used to predict the stock price,using Adam algorithm and improved Adam algorithm for comparative analysis of experiments.The experimental results show that,compared with the use of Adam algorithm,the accuracy of the improved Adam algorithm is increased from 57.7%to 65.8%in predicting the accuracy of stock fluctuations.The relative error(relative error ratio between absolute error and true value)is reduced from 3.98%to 2.78%.It can be seen that when using LSTM neural network,the improved Adam algorithm has better prediction results than Adam algorithm.