A comparison of different methods for combining multiple neural networks models

A single neural network model developed from a limited amount of data usually lacks robustness. Neural network model robustness can be enhanced by combining multiple neural networks. There are several approaches for combining neural networks. A comparison of these methods on three nonlinear dynamic system modelling case studies is carried out in this paper. It is shown that selective combination and combining networks of various structures generally improve model performance. The principal component regression approaches generally give quite consistent good performance.