Principal Component Analysis

Discussion in 'Technical Analysis' started by PoundTheRock, Oct 18, 2005.

  1. For stats gurus in the house, I have 5000 samples of a pattern (about a dozen input variables) and want to maximize the MFE. Should I do a factor analysis first, or is there a way to identify the clusters that maximize the MFE?
  2. Factor analysis is probably the way to go, but if you want to market the product you should re-express the statistical factors in terms of something the client can relate to.
  3. eitanhir


    pca is good for reducing the number of dimensions. i don't think this is what you need (maybe i don't understand your situation). i think using kmeans is a better option - it would help you find clusters in your 12 dimensions space. each cluster can have different mean result.
  4. Thanks guys -- perfect. PCA for eliminating some of the redundancy and kmeans for the clusters.