A dynamic attribute reduction algorithm based on compound attribute measure

Abstract

Attribute measure plays a vital role in the process of attribute reduction in decision systems. In spite of many attribute measures in heuristic attribute reduction algorithms can well evaluate the quality of attributes in decision systems, they do not consider the significance of information granularity beyond the positive region, such that some useful information not in the positive region may be loss in determining attribute quality. In addition, the attributes of decision systems usually vary dynamically with time in the real-world, correspondingly, attribute reduction needs updating to acquire new attribute reduct. In this paper, we firstly put forward a new compound attribute measure, which not only considers the measures of certain information in the positive region, but also considers the differences of information granularity of each attribute. Then based on the proposed compound attribute measure, we develop a dynamic attribute reduction algorithm for new reduct computation in dynamic decision systems. A case study is to illustrate the proposed reduction algorithm based on the compound attribute measure can find more useful attributes to guide the search for the best attribute reduct.