view article
Figure 4
An illustration of the problems with principal components analysis and other variance-based methods. The data could be cleanly separated into two clusters by the direction labelled y′, as obtained from linear discriminant analysis (LDA). However, projections along the principal component direction (y) would not produce any such clustering.

Journal logoBIOLOGICAL
CRYSTALLOGRAPHY
ISSN: 1399-0047
Volume 60| Part 12| November 2004| Pages 2133-2143
doi:10.1107/S090744490402061X
Follow Acta Cryst. D
Sign up for e-alerts
E-alerts
Follow Acta Cryst. on Twitter
Twitter
Follow us on facebook
Facebook
Sign up for RSS feeds
RSS