A Hybrid Genetic XK-means++ Clustering Algorithm with Empty Cluster Reassignment

Abstract

K-means is a classical clustering algorithm in many research areas, such as, document clustering, bioinformatics, image segmentation and pattern recognition. But, K-means is sensitive to the initial choice of cluster centers. A successful modification of K-means has been introduced in the literature by improving arbitrary cluster centers in the initialization stage-called K-means++. eXploratory K-means(XK-means)is another modification of K-means, which added an exploratory disturbance onto the vector of cluster centers, so as to improve the condition of sensitivity to the initial centers and jump out of the local optimum. However, the empty clusters may appear in the process of XK-means. The efficiency of the clustering result will be damaged by these empty clusters. In this paper, we try adding exploratory disturbance in K-means++ referred to as XK-means++. The same as XK-means, empty clusters also appear in the iteration process of XK-means++. Therefore, in this paper, an empty-cluster-reassignment technique is introduced and used in XK-means++(called EXK-means++). Furthermore, we combined the EXK-means++ with genetic mechanism, obtain a GEXK-means++ clustering algorithm. The data simulation results show that GEXK-means++is promising and effective.