A NEW MEASURE OF SIGNIFICANCE OF CONDITION ATTRIBUTE AND ITS USE IN ATTRIBUTE REDUCTION

摘要

In Pawlak's rough set system, approximation quality can be used to measure the classification capability of a condition attribute and can be used to define the significance of condition attributes. But the measure only gives us the determinate classification capability, and does not give us the uncertain classification capability. The information entropy is a mean value in terms of probability that embodies the whole information of an attribute, measuring the mean classification capability. The definition of mutual information is not intuitionistic and not easily understood; the calculation of the mutual information is somewhat complicated too. In this paper, we give a definition of probability equivalence between condition and decision attributes, and based on the probability equivalence we give a measure of significance of a condition attribute. An Illustrative example shows that with it the optimal attribute reduction can be gotten.