The optimal structure of recurrent neural networks for forecasting

A recurrent neural network is a network with feedback (closed loop) connections. The structure of the network seriously affects the performance of the network model. As the number of nodes in the input and output layers are application-dependent, the optimal structure problem reduces to the problem of how to optimally choose the number of hidden nodes in the hidden layer. By combining the Baum-Haussler rule and the Bayesian Information Criterion, we come up with a new rule to obtain the desired structure. By an empirical study, using the rainfall and discharge data to forecast the daily discharges at two important stations, we found that the proposed rule works satisfactory.