Density based Projection Pursuit Clustering

Clustering of high dimensional data is a very important task in Data Mining. In dealing with such data, we typically need to use methods like Principal Component Analysis and Projection Pursuit, to find interesting lower dimensional directions to project the data and hence reduce their dimensionalit...

Full description

Saved in:
Bibliographic Details
Main Authors: Tasoulis, S. K., Epitropakis, M. G., Plagianakos, V. P., Tasoulis, D. K.
Format: Conference Proceeding
Language:English
Subjects:
Clustering algorithms
Electronic mail
Machine learning algorithms
Optimization
Partitioning algorithms
Principal component analysis
Vectors
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Clustering of high dimensional data is a very important task in Data Mining. In dealing with such data, we typically need to use methods like Principal Component Analysis and Projection Pursuit, to find interesting lower dimensional directions to project the data and hence reduce their dimensionality in a manageable size. In this work, we propose a new criterion of direction interestingness, which incorporates information from the density of the projected data. Subsequently, we utilize the Differential Evolution algorithm to perform optimization over the space of the projections and hence construct a new hierarchical clustering algorithmic scheme. The new algorithm shows promising performance over a series of real and simulated data.
ISSN:1089-778X
1941-0026
DOI:10.1109/CEC.2012.6253006