Research on the attribute reduction method based on the best approximation set

Abstract

How to obtain fewer attribute sets while keeping some features of the data system unchanged is a widespread research interest in academia. In this paper, the following work is done based on the utility bias (abbreviated as w-UB) of the target set: 1) The best approximation of multiple objective sets is discussed, and the metric model in the sense of w-UB is also given. 2) Taking the best lower (upper) approximation set of the decision class as the core indicator, a best approximation-based attribute reduction method (abbreviated as δ⊕w-BAR) is proposed. 3) The relationship between δ⊕w-BAR and several commonly used reduction methods is discussed, further the performance of δ⊕w-BAR is analyzed from different perspectives. Theoretical analysis and simulation experiments show that δ⊕w-BAR has good interpretability and can briefly integrate decision preference into the data decision-making process. So it has good application value.