Automatic Discovery of Clusters by Removing Noisy Data

The key issue of determining a suitable (feasible) number of clusters still remains open. This paper proposes a graph-theoretic clustering iterative algorithm that employs a novel idea of using noise and information associated with it to determine clusters. The proposed method does not require any parameters whose values have to be supplied by the user. A series of experiments reported in the study show that the proposed algorithm can extract significant cluster information even in case of complicated geometry of data sets.