Clustering By Impact, Algorithms And Applications

George Mason University, Fairfax VA

Investigators

Abstract

The project will conduct research in the following topics: The design, implementation and testing of scalable method that uses self-similarity to cluster numerical data sets.The design, implementation and testing of a scalable method that uses entropy to cluster categorical data sets, based on the notion entropy.The design, implementation of testing of a scalable method of clustering mixed data sets.The research will have an impact on the following applications.Projective clustering: The project aim to use a combination of a clustering algorithms and singular value decomposition to discover the dimensions that are most relevant to each cluster in high dimensional data set.Tracking clusters in evolving data streams: The methods proposed are characterized by a concise representation of the clusters that have been found at a given point, and therefore allow the clustering of continuously incoming data streams. Labeling data through clustering: The incremental methods developed in the project to obtain new training data for classification algorithms.