Performance Evaluation of Spectral Clustering Algorithm using Various Clustering Validity Indices

M. T. Somashekara; D. Manjunatha

In spite of the popularity of spectral clustering algorithm, the evaluation procedures are still in developmental stage. In this article, we have taken benchmarking IRIS dataset for performing comparative study of twelve indices for evaluating spectral clustering algorithm. The results of the spectral clustering technique were also compared with k-mean algorithm. The validity of the indices was also verified with accuracy and (Normalized Mutual Information) NMI score. Spectral clustering algorithm showed better results when compared to k-mean algorithm. All indices showed consistent results with spectral clustering technique. Silhouette Index, Hartigan Index, Davies-Bouldin (DB) index and Krzanowski-Lai (KL) index failed to evaluate k-mean clustering. Surprisingly, all eleven indices showed acceptable results for spectral clustering algorithm. This article confirms the superiority of spectral clustering algorithm and also confirms that all 12 indices are suitable for evaluating spectral clustering.

Select Volume / Issues:

• Volume 5,Issue 6, Nov. 2014

Year:

2014

Type of Publication:

Article

Keywords:

Spectral Clustering; Validity Indices; NMI Score; K-Mean Algorithm

Journal:

IJECCE

Volume:

5

Number:

6

Pages:

1274-1276

Month:

November