Abstract
For data sets with nominal attributes, the discernibility based dissimilarity between clusters has been proposed. The dissimilarity is defined by the number of attribute subsets which can distinguish different clusters. Discernibility on an attribute subset implies that on its supersets. Therefore, when two clusters are discernible on a small attribute subset, those dissimilarity is small no matter how large differencese between objects in different clusters are. In this paper, we propose a new discernibly based dissimilarity reflecting differences between objects in different clusters.
