ABSTRACT

Clustering is the problem of discovering “meaningful” groups in given data. The first and common step in the process of Partitional Clustering is to decide the best value of K, the number of partitions. The clustering solution varies with K. Instead of clustering the data by guessing K value, in this paper we propose to cluster the data based on their similarity to obtain more meaningful clusters. Other characteristics of our clustering approach are (1) It deals with outliers (2) It deals the problem of clustering heterogeneous data (3) It reduces the high dimensionality of the term document matrix (4) It outperforms in accuracy the well- known clustering algorithm K-Means.