Exploring the Impact of Optimal Clusters on Cluster Purity

Machine Learning has been predominantly applied in several domains. Some automatic methods have been created using supervised techniques which require labels. However in quite a lot of real time scenarios, labels may not be available. In such scenarios, unsupervised methods play an important role. Clustering is popular unsupervised method where it groups objects with similar characteristics into a single cluster. However estimating the number of clusters is a critical task in clustering. If optimal number of clusters are not chosen, it may adversely affect clustering analysis. In this paper, we have analyzed experimentally the impact of number of clusters on cluster purity. Our experiments demonstrated that cluster purity varies up to 40-60% when optimal clusters are not chosen.