Service Analytics

6.Principal Component Analysis

Orthogonalization and Dimensionality Reduction

  • Orthogonalization alone has already some benefits: e. g. yields well conditionedmodels (recall last lecture).
  • But sorting the orthogonal vectors (principal components) in the order of descending variance establishes a basis for dimensionality reduction.
  • Since the variance contribution gets smaller and smaller from one principal component to the next, throwing away the least significant components does not result in a substantial error.

Diskussion