A positive region-based dimensionality reduction from high dimensional data

Abstract

Dimensionality reduction is an important attribute process work. Dimensionality reduction, i.e, attribute reduction is to delete some uncesserary attributes at rough sets. At present, many attribute reduction methods have provided to delete some superfluous and irrelevant attributes from large-scale complete data sets. The main drawback of most attribute reduction algorithms is that they can not remove some examples in the process of dimensionality reduction, which degrades a computational efficiency of attribute reduction. To overcome this drawback, an improved attribute reduction algorithm for complete data sets is proposed. In addition, the classification performance of attribute reduction is optimized. At first, the compact decision system is presented to delete some repeated objects. Then the significance measure of attributes is provided for candidate attributes. In addition, the novel approach of attribute reduction under the proposed significance measure of attributes was developed. In order to verify the efficiency of our given algorithm, the experiments on UCI datasets are performed by comparing with other attribute reduction algorithms. The results on the experiments tell us that our given algorithm obtains promising improvement for selecting an attribute reduct.