抽象的な

NOVEL INITIALIZATION TECHNIQUE FOR K-MEANS CLUSTERING USING SPECTRAL CONSTRAINT PROTOTYPE

Mrs.S. Sujatha and Mrs. A. Shanthi Sona

Abstract---Clustering is a general technique used to classify collection of data into groups of related objects. One of the most commonly used clustering techniques in practice is K-Means clustering. The major limitation in K-Means is its initialization technique. Several attempts have been made by many researchers to solve this particular issue, but still there is no effective technique available for better initialization in K-Means. In general, K-Means follows randomly generated initial starting points which often result in poor clustering results. The better clustering results of K-Means technique can be accomplished after several iterations. However, it is very complicated to decide the computation limit for obtaining better results. In this paper, a novel approach is proposed for better initialization technique for K-Means using Spectral Constraint Prototype (K-Means using SCP). The proposed method incorporates constraints as vertices. In order to incorporate the constraints as vertices, SCP approach is used. The proposed approach is tested on the UCI Machine learning repository. The proposed initialization provides better clustering accuracy with lesser execution time.