Cluster Validity Analysis on Soft Set Based Clustering

Abstract

The issue of data uncertainties are very important in categorical data clustering since the boundary between created clusters are very arguable. Therefore the algorithm called Maximum Attribute Relative (MAR) that is based on the attribute relative of soft-set theory was proposed previously. MAR exploiting the data uncertainties in multi-value information system by introducing a series of clustering attribute. The clusters will be form by using this selected clustering attributes. However, clustering algorithm define clusters that are not-known a priori. Hence, the final clusters of data requires some validation techniques. In this paper, the validity of the clusters produced by MAR was evaluated. The two datasets obtained from UCI-ML repository and an examination results obtained from Malaysian Ministry of Education. The results shows that the clusters produced by MAR has objects similarity up to 99%.