Improved K-means clustering on Hadoop

Abstract

Clustering is the portioning method in which we grouped similar attribute items. Recently data grows rapidly so data analysis using clustering getting difficult. K-means is traditional clustering method. K-means is easy to implement and scalable but it suffers from local minima and sensitive to initial cluster centroids. Particle swarm optimization is mimic behavior based clustering algorithm based on particle’s velocity but it suffers from number of iterations. So we use PSO for finding initial cluster center and then use this centroids for K-means clustering which is running parallel on Hadoop. Hadoop is used for large database. We try to find global clusters in limited iterations.