A simulated-annealing-based approach for simultaneous parameter optimization and feature selection of back-propagation networks

Abstract

The back-propagation network (BPN) can be used in various fields. Nevertheless, different problems may require different parameter settings for network architectures. Rule of thumb or ''trial and error'' methods are usually used to determine them. However, these methods may lead worse parameter settings for network architectures. On the other hand, although a dataset may contain many features, not all features are beneficial for classification in BPN. Therefore, a simulated-annealing-based approach, denoted as SA+BPN, is proposed to obtain the optimal parameter settings for network architectures of BPN, and to select the beneficial subset of features which result in a better classification. In order to evaluate the proposed SA+BPN approach, datasets in UCI Machine Learning Repository are used to evaluate the performance of the proposed approach. The experimental results show that the parameter settings for network architectures obtained by the proposed approach are better than those of other approaches. When the feature selection is taken into consideration, the classification accuracy rates of most datasets are increased. Therefore, the developed approach can be utilized to find out the optimal parameter settings for network architectures of BPN, and discover the useful features effectively.