An Effective Approach for Selecting Cluster Centroids for the k-means Algorithm using IABC Approach

Abstract

K-means is a popular grouping technique for unsupervised data. Though the technique is simple and potential it suffers from the ambiguity in the selection of k (number of clusters). The selection or initialization of k-cluster centroids significantly impacts the performance of the k-means. The distribution of data with the possibility of outliers may degrade the clustering strength. If the initialization of clustering is best fitted then the rest of the k-means process will save the execution time and can offer better clustering. A searching procedure is needed that learn the data under hand and decide the cluster initialization accordingly. ABC is a good technique for optimized search. In this paper, a study of different variants of ABC is undertaken in order to propose a new methodology named IABC. The methodology made use of bee intelligence for search optimization along with k-means and k-nearest neighborhood algorithms. The proposed approach adds the strength to the k-means approach in terms of optimum centroid initialization. Experiments are done on two popular datasets to compare the proposed approach with the existing techniques. The comparisons are made between the existing and proposed methods. Better and encouraging results are found. The proposed process is able to save a significant amount of time and can offer better and accurate clustering.