VatAna-package: Visual Assessment of Clustering Tendency for Finding the...

In zcebeci/VatAna: Visual Assessment of Clustering Tendency for Finding the Number of Clusters in a Dataset

Description

The package VatAna provides the functions related to the Visual Assessment of Cluster Tendency (VAT) algorithm proposed by Bezdek & Hathaway (2002).

Details

The partitioning algorithms require a priori estimate of number of clusters (k) as an input parameter, and thus the success of partitioning depends mostly on this parameter. In order to find an optimal estimation of k, the obtained results are checked by the cluster validity indices at the end of each run of successive cluster analyses. Unfortunately, this kind of cluster validation is time consuming task, and also depends on the clustering indices which may not guarantee the quality of clustering since their performances vary with complexity in data structures (Cebeci, Kavlak & Yildiz, 2017). In order to find an optimal number of clusters in datasets alternatively, one also can benefit from the visual techniques like visual assessment of clustering tendency (VAT) algorithm for finding the k in order to start a clustering session. The VAT is a frontier algorithm which produces a gray-level image of reordered distance matrix showing existing clustering with dark rectangular blocks along the diagonal of it.

The function VatAna::vatimage of the package builds the ordered dissimilarities matrix (ODM) to be displayed as an image (ODI) for visual inspection as introduced in Bezdek & Hathaway (2002).

The function VatAna::vatdisp displays all or any of dissimilarities images such as DM, ODM or BDM.

The function VatAna::greyimage generates and displays a greyscale dissimilarities matrix.

The function VatAna::binimage generates and displays a binary dissimilarities matrix. The argument method is used to assign a thresholding method such as 'otsu', 'auto', 'median', 'fixed' and 'quant'.

The function VatAna::findk counts the dark blocks by moving the edges of them as described by Cebeci and Yildiz (2015). The counted number of blocks is proposed as the number of clusters in the examined dataset.

The function VatAna::findk2 counts the dark blocks by moving the diagonal of ODM of them as described by Cebeci and Yildiz (2015). This function can not discrimnate the overlapped blocks and finds less number of blocks than the function findk. The counted number of blocks is proposed as the number of clusters for the examined dataset.

For testing purposes, the function VatAna::genbinimg generates the artificial binary images containing a number of user-defined blocks.

The package also contains a dataset named VatAna::protcarbo, has 3 actual clusters with 2 variables and 15 rows. It can be used for testing the success to find the optimal numbcer of cluster using different threshold levels in generation of binary image matrices.

For finding the optimal number of clusters in a dataset: a) First run VatAna::vatimage with the dataset. b) In order to get a proposal for the number of clusters in the dataset run VatAna::findk and/or VatAna::findk2 using ODM from the result of previous step. c) In order for visual inspection for VAT, draw the images of dissimilarities matrices using the function VatAna::vatdisp.

Author(s)

Zeynel Cebeci

References

Bezdek, J. C., & Hathaway, R. J. (2002). VAT: A tool for visual assessment of (cluster) tendency. In Proceedings of the 2002 International Joint Conference on Neural Networks. IJCNN'02 (Cat. No. 02CH37290), Vol. 3, pp. 2225-2230. IEEE.

Cebeci, Z. & Yildiz, F. (2015). Gorsel Kumelenme Egilimi Degerlendirmesi ve R'de Uygulamasi. Cukurova Universitesi Ziraat Fakultesi Dergisi, 30 (2), 1-8. (URL: https://dergipark.org.tr/en/download/article-file/219860)

Cebeci, Z., Kavlak, A. T. & Yildiz, F. (2017). Validation of fuzzy and possibilistic clustering results. In International Artificial Intelligence and Data Processing Symposium (IDAP 2017), pp. 1-7. <doi:10.1109/IDAP.2017.8090183>. IEEE.

See Also

binimage, findk, findk2, genbinimg, greyimage, protcarbo, vatimage, vatdisp

zcebeci/VatAna documentation built on Dec. 25, 2019, 7:07 p.m.