Correlation of Attributes in Random Decision Information Systems and Its Application to Attributes Reduction

Abstract

The conditional information entropy of a condition attributes set with respect to another condition attributes set is defined in a random decision information system, by using the conditional information entropy the correlation of a conditional attributes set relative to another condition attributes set with respect to decision attributes set, the properties of correlation are discussed, and some concepts about random decision information system, such as necessary attribute, consistent attribute and attributes nuclear, etc, are described by using the correlation. By using the correlation of attributes an attribute reduction method and an attribute reduction algorithm are proposed respectively, and an example shows that the algorithm is effective.