Comparison of pattern extraction capability between self-organizing maps and principal component analysis

Yoshihiko Iseri, Tomonori Matsuura, Satoshi Iizuka, Koji Nishiyama, Kenji Jinno

Research output: Contribution to journalArticle

10 Citations (Scopus)

Abstract

Principal Component Analysis (PCA) has been extensively used for multivariate data analysis including climate data analysis. However, recent studies suggest the possible use of Self-Organizing Maps (SOM) for climate data analysis. In order to adequately utilize the PCA and SOM, it is crucial to clarify the difference of both methods. This study compared the pattern extraction capability of SOM and PCA in order to clarify the difference of both methods. Comparisons of the methods were performed by conducting two kinds of simulations. The first stimulation confirmed SOM can extract non-orthogonal patterns while the patterns extracted by PCA are mutually orthogonal. The second simulation investigated how the patterns extracted by each method would change when new inputs were added. When new inputs were added, PCA did not extract the pattern in the original inputs and additional inputs. On the other hand, SOM extracted the pattern in original and additional inputs.

Original languageEnglish
Pages (from-to)37-47
Number of pages11
JournalMemoirs of the Faculty of Engineering, Kyushu University
Volume69
Issue number2
Publication statusPublished - Jun 1 2009
Externally publishedYes

All Science Journal Classification (ASJC) codes

  • Energy(all)
  • Atmospheric Science
  • Earth and Planetary Sciences(all)
  • Management of Technology and Innovation

Fingerprint Dive into the research topics of 'Comparison of pattern extraction capability between self-organizing maps and principal component analysis'. Together they form a unique fingerprint.

  • Cite this