Cascade Forward Neural Network for Time Series Prediction

Abstract

Cascade-forward neural network is a class of neural network which is similar to feed-forward networks, but include a connection from the input and every previous layer to following layers. In a network which has three layers, the output layer is also connected directly with the input layer beside with hidden layer. As with feed-forward networks, a two-or more layer cascade-network can learn any finite input-output relationship arbitrarily well given enough hidden neurons. Cascade-forward neural network can be used for any kind of input to output mapping. The advantage of this method is that it accommodates the nonlinear relationship between input and output by not eliminating the linear relationship between the two. In this study, we apply the network in time series field. The optimal architecture was determined computationally by using incremental search method in both input and hidden units. The simple one was built first, and then the more complex is constructed by adding the units one by one. The optimal one is chosen then by using the mean square error criteria.