Principal Component Analysis (PCA), also known as the Karhunen-Loeve or Hotelling transform, provides an elegant method for achieving reduction in dimensionality of a data set. The reduced data set can then be used as input to another analysis, e.g., involving artificial neural nets. Sometimes, the principal components are an end in themselves and may be subjected to interpretation. PCA identifies linear combinations of raw parameters accounting for maximum variance in a data set. A compression of the data is obtained by ignoring those components that represent the least variance in the data. In this paper, we describe the transform which reduces a general n*m data array into an n*p array (where p << m) and review some applications of the principal component analysis with special emphasis on classification of stellar spectra.
in Automated Data Analysis in Astronomy,
R. Gupta, H.P. Singh, C.A.L. Bailer-Jones (eds.), Narosa Publishing House, New Delhi,
India, 2001
[PDF version] 757Kb, 14 pages