An enhanced visual approach for accessing the clustering tendency of big data

Abstract

Cluster analysis aims to create the groups for the data objects based on the assessment of similarity features. It is an essential unsupervised technique for the unlabelled datasets. For example, data clustering methods' primary problem is that k-means suffer from the intractable assignment of 'k' value by external interference (or user). Finding the number of clusters 'k' is called a clustering tendency. Existing visual approaches, i.e., visual access tendency (VAT), cosine-based VAT (cVAT), cosine-based spectral VAT(CS-VAT), are suitable for determining the value of cluster tendency of regular data. The Clustering using Improved Visual Assessment of Tendency (ClusiVAT) performs as the best for significant data clustering than other visual approaches. It uses the sampling technique for faster results; however, it perfectly works for Gaussian-based generated datasets. Thus, the proposed work develops the enhanced visual approaches for obtaining the quality of clusters for the typical datasets. Performance of enhanced visual approaches is demonstrated in the experimental study using benchmarked datasets.