Input selection based on an ensemble

NEUROCOMPUTING

卷:34

Article

van de Laar, P ; Heskes, T

关键词: architecture selection;    combining classifiers;    combining predictors;    feature selection;    input selection;    knowledge extraction;    subset selection;   

DOI  :  10.1016/S0925-2312(00)00294-0

来源: Elsevier

【 摘 要 】

Since an ensemble of neural networks outperforms a single network, we expect that the selection of input variables based on an ensemble is superior to the selection based on a single neural network. In this article, we will present an algorithm that performs input selection based on an ensemble of neural networks. Using this algorithm, the correct sets of variables were found for two artificial problems. Furthermore, for two real-world problems, we determined the relevance of the input variables. Our predictions were equal or better than the predictions of other methods described in the literature. (C) 2000 Elsevier Science B.V. All rights reserved.

【 授权许可】

Free