Open AccessChoice of the Number of Hidden Layers for Back Propagation Neural Network Driven by Stock Price Data and Application to Price Prediction
Author(s) -
Peipei Zhang,
Chuanhe Shen
Publication year - 2019
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1302/2/022017
Subject(s) - artificial neural network , computer science , stock price , stock market , stock market prediction , predictive modelling , stock (firearms) , nonlinear system , construct (python library) , backpropagation , artificial intelligence , data mining , econometrics , machine learning , mathematics , engineering , paleontology , series (stratigraphy) , mechanical engineering , physics , horse , quantum mechanics , biology , programming language
Since the stock market is dynamic and nonlinear, we adopt the neural network to forecast the stock price. We construct the single hidden layer prediction model firstly, and analyse the effect of prediction accuracy on neurons amount and epochs. To improve the prediction accuracy and operating rate, we then construct the multiple hidden layers prediction model, and provide some theory guide on setting the number of each hidden layer for neural network with multiple hidden layers. Finally, we make a choice of the number of hidden layers by analysing the effect of stock price prediction, and the empirical results obtained demonstrate that the prediction performance of two hidden layers prediction model is better than that of the single hidden layer prediction model. Additionally, the empirical results obtained also demonstrate that the more epochs of training network, the better the results obtained with using the same number of neurons.