A novel clustering framework using farthest neighbour approach

In this digital world, we are facing the flood of data, but depriving for knowledge. The eminent need of mining is useful to extract the hidden pattern from the wide availability of vast amount of data. Clustering is one such useful mining tool to handle this unfavorable situation by carrying out crucial steps refers as cluster analysis. It is the process of a grouping of patterns into clusters based on similarity. Partition based clustering algorithms are widely accepted for much diverse application such as pattern analysis, image segmentation, identification system. Among the different variations of the partition based clustering, due to its monotony and ease of implementation K-means algorithm gained a lot of attraction in the various field of research. A severe problem associated with the algorithm is that it is highly sophisticated while selecting the initial centroid and may converge to a local optimum solution of the criterion function value if the initial centroid is not chosen accurately. Additionally, it requires the prior information regarding a number of clusters to be formed and the computation of K-means are expensive. K-means algorithm is a two-step process includes initialization and assignment step. This paper works on initialization step of the algorithm and proposed an efficient enhanced K-means clustering algorithm which eliminates the deficiency of the existing one. A new initialization approach has been introduced in the paper to drawn an initial cluster centers for K means Algorithm. The paper also compares proposed technique with K-means technique.