Author : Barakbah, Ali Ridho; Helen, Afrida;
Performance of K-means algorithm which depends highly on initial starting points can be trapped in local minima and led to incorrect clustering results. The lack of K-means algorithm that generates the initial centroids randomly does not consider the placement of them spreading in the feature space. In this paper we propose a new approach to optimize the initial centroids for K-means. This approach spreads the initial centroids in the feature space so that the distance among them are as far as possible. Started from the center of the data, this approach chooses each initial centroids those reside in distant position among them. The experimental results show the improved solution using the proposed approach.
Keyword : clustering; K-means algorithm; initial centroids
Sumber : http://repository.petra.ac.id/65/