2

After running PCA on my data set, I noticed that using the three first eigenvectors, a separation between two different classes is still achievable (doing PCA on data from two classes). Unfortunately, it doesn't scale up very well and when I feed in data from 15 different classes, I get a big blob of dots when I plot the reprojected data in three dimensions. I want to find out if those classes are actually separated in higher dimensions. Is there techniques used to visualize scatter plots of data in more than 3 dimensions?

I have read about using different shapes (dots, triangles, squares, and their respective area) to fit more information in a 3D scatter plot but how would you code that relative to the values delivered by PCA?

amoeba
  • 93,463
  • 28
  • 275
  • 317
Mehdi
  • 161
  • 1
  • 3
  • First, you can plot all possible 2d scatterplots ("scatterplot matrix"), like e.g. in this answer http://stats.stackexchange.com/a/101490. Second, you can try to use a supervised method such as LDA, see here http://stats.stackexchange.com/questions/161362. – amoeba Jul 17 '15 at 10:00

0 Answers0