A Clustering-Oriented Star Coordinate Translation Method for Reliable Clustering Parameterization

Abstract

When conducting a clustering process, users are generally concerned whether the clustering result is reliable enough to reflect the actual clustering phenomenon. The number of clusters and initial cluster centers are two critical parameters that influence the reliability of clustering results highly. We propose a Clustering-Oriented Star Coordinate Translation (COSCT) method to help users determining the two parameters more confidently. Through COSCT all objects from a multi-dimensional space are adaptively translated to a 2D star-coordinate plane, so that the clustering parameterization can be easily conducted by observing the clustering phenomenon in the plane. To enhance the cluster-displaying quality of the star-coordinate plane, the feature weighting and coordinate arrangement procedures are developed. The effectiveness of the COSCT method is demonstrated using a set of experiments.