Publication View

Entropy in Multivariate Analysis: Projection Pursuit GPNason. School of Mathematical Sciences, University of Bath, (2008)

Abstract
Projection pursuit is an exploratory data-analytic method in multivariate (MV) analysis. It is similar to the well-known principal components analysis (PCA) in that it can be used to nd interesting structure within a MV data set. However, unlike PCA, which nds linear projections of maximum variance, projection pursuit nds linear projections of maximum non-normality, which sometimes is better at revealing structure within a MV data set. We describe how the negative Shannon entropy can be used for measuring non-normality. As a result we can view projection pursuit with the Shannon index as a method which nds the projection with the maximum entropy. We outline the constrained optimising projection pursuit algorithm and mention brie y the role of sphering a MV data set. Finally we illustrate the method as applied to the famous Lubischew beetle data and mention how it can be applied to multispectral images. 1

Publication details
Download http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.64.4535
Source http://www.stats.bris.ac.uk/~magpn/Research/papers/EntInMVA.pdf
Contributors CiteSeerX
Repository CiteSeerX - Scientific Literature Digital Library and Search Engine (United States)
Type text
Language English