Cluster Tendency Methods for Visualizing the Data Partitions

Abstract

Clustering is widely used technique for grouping of data objects based on similarity features. The similarity features are derived from the similarity or dissimilarity metrics like Euclidean, cosine etc. Traditional clustering methods such as k-means, and other graph-based techniques are major techniques for discovery of clusters. However, these methods require user interference for determining the number of clusters initially. Determining the number of clusters for given data is known as cluster tendency. There is chance for getting poor clustering results when using either k-means or graph-based clustering methods with intractable value of ‘k’ by user. Thus, it is required to focus on cluster tendency methods for knowing prior knowledge about number of clusters in clustering. This paper presents the various visual access tendency (VAT) methods for good assessment of number of clusters.