A unified evolutionary training scheme for single and ensemble of feedforward neural network

Artificial neural networks (ANNs) have been successfully applied to many areas. The key of success is to properly tune the architecture and the connection weights of the ANN and to have the trained ANN own higher generalization ability. For complicated problems, artificial neural network ensemble classifier, instead of a single ANN classifier, is considered. But, it is not so straight forward to construct the ANN ensemble. In this paper, we propose a unified evolutionary training scheme (UETS) which can either train a generalized feedforward neural network or construct an ANN ensemble. The performance of the UETS was evaluated by applying it to solve the n-bit parity problem and the classification problems on five datasets from the UCI machine learning repository. By comparing with the previous studies, the experimental results reveal that the neural networks and the ensembles trained by the UETS have very good classification ability for unseen cases.