Determination the Number of Hidden Nodes of Recurrent Neural Networks for River Flow and Stock Price Forecasting
Suwarin Pattamavorakun, Suwat Pattamavorakun · 2007
The classic approach to time series forecasting is to undertake an analysis of the time series data. Recurrent neural networks (RNNs) are designed to learn sequential or time-varying patterns. Due to their dynamic nature, so RNNs are suitable for time series forecasting. As the number of nodes in the input and output layers are application - dependent, the problem reduces to how to optimally choose the number of hidden nodes. This is achieved by using the Bayesian Information Criterion via the number of hidden nodes of recurrent neural network. It is also proved that there is a close link between the Bayesian Information Criterion and Efficiency Index, which indicates how good the model when it is used for the river flow and stock price data set.