An Efficient Data Pre-processing Procedure for Support Vector Clustering

Findings of  Dr. K. Raghava Rao, Y.A.Siva Prasad, B.Venugopal on "An Efficient Data Pre-processing Procedure for Support Vector Clustering" have been selected for NATIONAL CONFERENCE on "Imaging, Computing, Object and Mining (ICOM'10)" Organized by Department of Computer Science and Engineering ANNA UNIVERSITY, TIRUCHIRAPPALLI scheduled on May 7th and 8th 2010 at Anna University Tiruchirappalli,TAMILNADU

Abstract:

This paper presents an efficient data pre-processing procedure for the support of vector clustering (SVC) to reduce the size of a training dataset. Solvingthe optimization  problem and labeling the data points with cluster labels are time-consuming in the SVC training procedure. This makes using SVC to process largedatasets inefficient. We proposed a data pre-processing procedure to solve the problem.  The procedure contains a shared nearest neighbour (SNN) algorithm, andutilizes the concept of unit vectors for eliminating insignificant data points from the dataset. Computer simulations have been conducted on artificial and benchmarkdatasets to demonstrate the effectiveness of the proposed method.

Keywords? Support Vector Clustering, Shared Nearest Neighbours, Noise Elimination.