NDER Attribute Reduction via an Ensemble Approach

摘要

Traditional attribute reduction based on neighborhood decision error rate aims to reduce the decision errors through selecting valuable attributes. To further improve the performances of the selected attributes in reducts, an ensemble selector is introduced into such framework. Different from the previous strategy, our approach is realized through considering a set of the fitness functions instead of one and only one fitness function, which makes the ensemble selecting of attribute is possible. The experimental results on 10 UCI data sets and 2 KEEL data sets demonstrate that our ensemble selector is effective in improving the stabilities of both reducts and classification results. In addition, the classification accuracies can also be increased.