SDCC: A New Stable Double-Centroid Clustering Technique Based on K-Means for Non-spherical Patterns

Abstract

Numerous existing partitioning clustering algorithms, such as K-means, are developed to discover clusters that fit some of the static models. These algorithms may fail if it chooses a set of incorrect parameters in the static model with respect to the objects being clustered, or when the objects consist of patterns that are of non-spherical or not the same size. Furthermore, they could produce an instable result. This investigation presents a new partition clustering algorithm named SDCC, which can improve the problem of instable results in partitioning-based clustering, such as K-means. As a hybrid approach that utilizes double-centroid concept, the proposed algorithm can eliminate the above-mentioned drawbacks to produce stable results while recognizing the non-spherical patterns and clusters that are not the same size. Experimental results illustrate that the new algorithm can identify non-spherical pattern correctly, and efficiently reduces the problem of long computational time when applying KGA and GKA. It also indicates that the proposed approach produces much smaller errors than K-means, KGA and GKA approaches in most cases examined herein.