Clustering Performance Analysis

Abstract

Clustering plays a significant role in identifying the intrinsic structure of data. In this paper, various clustering algorithms are compared on real, numerical, categorical datasets around the cluster size. From the analysis, it is inferred that the repetition of KMeans many times does not bring better significant iterations since it starts randomly. It purely depends on the initial choice of the centroid of clusters. The sum of squared error decreases with increasing cluster size. The Expectation–Maximization (EM) is time-consuming than KMeans.