A Propound Method for the Improvement of Cluster Quality
In this paper Knockout Refinement Algorithm (KRA) is proposed to refine original clusters obtained by applying SOM and K-Means clustering algorithms. KRA Algorithm is based on Contingency Table concepts. Metrics are computed for the Original and Refined Clusters. Quality of Original and Refined Clusters are compared in terms of metrics. The proposed algorithm (KRA) is tested in the educational domain and results show that it generates better quality clusters in terms of improved metric values.
Keywords: Web Usage Mining, K-Means, Self Organizing Maps, Knockout Refinement Algorithm (KRA), Davies Bouldin (DB) Index, Dunns Index, Precision, Recall, F-Measure
Download Full-Text
ABOUT THE AUTHORS
Shveta Kundra Bhatia
Shveta Kundra Bhatia is working as an Assistant Professor in the Department Of Computer Science, Swami Sharaddhanand College, University of Delhi. Her research area is Web Usage Mining and is currently pursuing PhD under Dr. V.S. Dixit from Department of Computer Science, University of Delhi.
V. S. Dixit
Dr. V. S. Dixit is working in the Department Of Computer Science, Atma Ram Sanatan Dharam College, University of Delhi. His research area is Queuing theory, Peer to Peer systems, Web Usage Mining and Web Recommender Systems. He is currently engaged in doing the research. He is Life member of IETE.
Shveta Kundra Bhatia
Shveta Kundra Bhatia is working as an Assistant Professor in the Department Of Computer Science, Swami Sharaddhanand College, University of Delhi. Her research area is Web Usage Mining and is currently pursuing PhD under Dr. V.S. Dixit from Department of Computer Science, University of Delhi.
V. S. Dixit
Dr. V. S. Dixit is working in the Department Of Computer Science, Atma Ram Sanatan Dharam College, University of Delhi. His research area is Queuing theory, Peer to Peer systems, Web Usage Mining and Web Recommender Systems. He is currently engaged in doing the research. He is Life member of IETE.