An Enhanced K-Medoid Clustering Algorithm

Abstract

Data mining is a technique of mining information from the raw data. It is a non trivial process of identifying valid and useful patterns in data. Some of the major Data Mining techniques used for analysis are Association, Classification and Clustering etc. Clustering is used to group homogenous kind of data, but it is different approach from classification process. In the classification process data is grouped on the predefined domains or subjects. A basic clustering technique represents a list of topics for each data and calculates the distance for how accurately a data fit into a group. The Cluster is helpful to get fascinating patterns and structures from an outsized set of knowledge. There are a lots of clustering algorithms that have been proposed and they can be divided as: partitional, grid, density, model and hierarchical based. This paper propose the new enhanced algorithm for k-medoid clustering algorithm which eliminates the deficiency of existing k-medoid algorithm. It first calculates the initial medoids ‘k’ as per needs of users and then gives relatively better cluster. It follows an organized way to generate initial medoid and applies an effective approach for allocation of data points into the clusters. It reduces the mean square error without sacrificing the execution time and memory use as compared to the existing k-medoid algorithm.