DefinePK

Abstract

The process of clustering is denoted as data grouping analysis that have a set of data objects and they can be defined as distinguished groups or subsets according to the similarity features; assessment of number of clusters is a key issue in data clustering problem. The popular k-means may also suffer from the issue of initial number of clusters ‘k’; thus, it is noted that k-means may outputs the poor clustering results if user attempts incorrect ‘k’ value. Determining the number of clusters for given set of data objects are called as cluster tendency. In k-means, assessing the ‘k’ value is cluster tendency for given data. This paper is majorly focused on how to perform effective assessment of cluster tendency for real-life datasets. Visual access tendency (VAT) is an optimal choice for assessment of clusters or cluster tendency. It finds the dissimilarity features using Euclidean metric and uses this matrix for determining the value of cluster tendency. Cosine metric is also most successful measure; however it computes the similarity between data objects using a single reference point. The proposed work uses multi view-points for measuring the accurate similarity matrix between data objects, and it is known as multi view-point based cosine similarity measure (MVS). The proposed MVS-VAT is experimented on various datasets for demonstrating better assessment of cluster results when compared to VAT and other related versions of VAT.