An Efficient Approach to Determine Number of Clusters Using Principal Component Analysis
V. P. Sree Divya, K. Nirmala Devi · 2018 International Conference on Current Trends towards Converging Technologies (ICCTCT) · 2018
Progressing efficient clustering technic for a high dimensional dataset is a challenging problem by reason of generating empty objects. In this paper uses a Kmeans clustering algorithm which is excellent for its easiness. However, the Kmeans method combines to one of many local minima. And it is noted that the final result depends on the initial centroid points (means). Many technics have been proposed to estimate the optimal number of clusters. In our proposed method, we have used silhouette with Principal Component Analysis (PCA) for empty cluster reduction and to find the new initial centroid for Kmeans. The proposed system uses various dataset such as iris, wine, thyroid, yeast and solar datasets (Ames, Chariton, Calmar stations). The results of the proposed estimation have better cluster estimation results while comparing to other estimation algorithms.