SPPS: Supervised Projected Clustering Method Based on Particle Swarm Optimization

Abstract

Supervised clustering algorithms are applied on classified examples with the goal of determining class-uniform clusters. These methods evaluate clustering solutions based on class impurity unlike traditional clustering methods. These methods can be used for tasks like data editing and learning of subclasses to enhance classification methods. Supervised clustering methods have been proposed in literature to find class-uniform full dimensional clusters. But for high dimensional dataset with subspace clusters there is need for supervised clustering method which finds class-uniform subspace clusters. In this paper we propose Supervised Projected clustering Particle Swarm optimization method (SPPS method). The proposed method has been applied on Wisconsin breast cancer data to find subspace clusters present in this dataset. The SPPS method may be used for pre-processing of high dimensional datasets with subspace clusters.