Attribute reduction based on fusion information entropy

Abstract

The attribute reduction algorithm based on fuzzy rough set theory is a commonly used method for dealing with uncertain information. However, many existing attribute reduction algorithms do not consider the information provided by the upper approximation. In fact, the upper approximation contains some certain decision information and decision risks brought by uncertain information. Therefore, we propose an attribute reduction algorithm based on fusion information entropy (FIE) to make proper use of the upper approximation. Firstly, we define the concepts of kernel decision degree and general decision degree, and flexibly combine them using the precision decision index. This method overcomes the limitations of existing attribute reduction methods in considering the upper approximation information. Then, we introduce the concept of fusion information entropy and measure the importance of attribute subsets by analyzing the changes in fusion information entropy. Finally, we design the corresponding attribute reduction algorithm. Extensive comparative experiments on public datasets show that the FIE algorithm based on fusion information entropy achieves higher classification accuracy while retaining fewer attributes compared with other state-of-the-art reduction algorithms. These results confirm the effectiveness and superiority of the FIE method proposed in this paper.