-
Student Participants
- Mindy Hong, Emory University
- Robert Pearce, NC State University
- Kevin Valakuzhy, University of North Carolina at Chapel Hill
- Advisors
- Carl D. Meyer (Faculty Advisor, NC State)
- Shaina Race (Graduate Student Advisor, NC State)
- Project Description
-
Data Mining is one of the fastest growing disciplines in mathematics and computer science today. Advances in data collection
and storage have allowed companies and scientific researchers to create huge stores of data in the hopes that data miners will
be able to discern valuable information from it. The vast majority of data mining models are examples of supervised learning;
a model is created using training and test data for which the variable to be predicted is known, and the goal is to minimize
the error of the prediction. We focused on unsupervised data mining techniques that aim to detect patterns and structure
in unlabeled data where no value for error or accuracy can be placed on the final result. Emphasis is on
clustering algorithms.
Many existing clustering algorithms are inadequate in that they
require knowledge of the number k of clusters that exist in the data, and in
that their underlying assumptions make them ineffective in certain situations. The work revolves around the
method of consensus clustering that seeks to rectify the latter problem by incorporating
the results of multiple clustering algorithms to achieve one final grouping. The goal is to
investigate a novel method of iterative consensus clustering (ICC) which solves both the problem of
determining the best value of k as well as improving cluster determination.
We show that iterating consensus clustering techniques widen the eigengap associated with the Perron cluster of eigenvalues and thus gives a more definitive and accurate estimation of the number of clusters.
-
Data Mining is one of the fastest growing disciplines in mathematics and computer science today. Advances in data collection
and storage have allowed companies and scientific researchers to create huge stores of data in the hopes that data miners will
be able to discern valuable information from it. The vast majority of data mining models are examples of supervised learning;
a model is created using training and test data for which the variable to be predicted is known, and the goal is to minimize
the error of the prediction. We focused on unsupervised data mining techniques that aim to detect patterns and structure
in unlabeled data where no value for error or accuracy can be placed on the final result. Emphasis is on
clustering algorithms.
Many existing clustering algorithms are inadequate in that they
require knowledge of the number k of clusters that exist in the data, and in
that their underlying assumptions make them ineffective in certain situations. The work revolves around the
method of consensus clustering that seeks to rectify the latter problem by incorporating
the results of multiple clustering algorithms to achieve one final grouping. The goal is to
investigate a novel method of iterative consensus clustering (ICC) which solves both the problem of
determining the best value of k as well as improving cluster determination.
-
Article and Poster Presentations
- The article Iterated Consensus Clustering: A Technique We Can All Agree On explains the details.
- A poster presentation was given at the Undergraduate Research Symposium, McKimmon Center, NC State University, August 2012