Title: Extending CLARANS to High-dimensional Spaces

When: Monday, February 23, 2004 12:00 noon

Where: IS Rm 503 (5th Floor Conference Hall)

Who: Denis Nkweteyim

Abstract: Cluster analysis is a data mining activity that aims at finding structure in data by grouping together, or clustering, data items that are highly similar to each other, and very dissimilar from data items in other clusters. Traditional approaches to clustering are limited by the fact that they are only efficient when dealing with small amounts of data, typically with very few dimensions. And although more recent approaches can deal with large datasets, most of them are limited to low dimensional data. In information retrieval and related domains, documents are usually represented in high-dimensional spaces and clustering approaches that deal with high dimensional data would be useful.

In this talk, I will present the results of a simulation study that shows that the CLARANS clustering algorithm traditionally used to cluster low dimensional data can be extended cluster high dimensional data too. Comparisons will be made with the PAM and CLARA algorithms, which form the basis of CLARANS.

Speaker Bio: Denis Nkweteyim is a Ph.D. candidate in the Information Science program of the School of Information Sciences, University of Pittsburgh. He holds a BS degree in Electrical & Electronic Engineering from the University of Wales, Cardiff, and an MS degree in Digital Communication Systems from Loughborough University. His current research interests are in the area of personalization systems.