1
$\begingroup$

I have a dataset where I need to explore using unsupervised technics (clustering and association rules). What are the best strategies to discretize the numeric attributes? Also, does this (attribute discretization) makes sense for clustering (using K-Means)?

Since I'm using weka, I know that I can just use the Discretize filter, with findNumBins option or useEqualFrequency. But are there other strategies?

Also, does it makes sense to use discretization on the attributes, and then look for clusters (using K-Means) or not? (I suspect that it doesn't make that much sense, but I just want to confirm)

Thanks!

$\endgroup$
2
$\begingroup$

Most clustering algorithms work best with continuous values (in particular k-means). Many cannot use categorial attributes and won't work very well with one-hot encodings either.

$\endgroup$
2
  • $\begingroup$ Thanks, and about for EM? Also what would be a good justification for that (The K-Means part)? $\endgroup$
    – Luis Alves
    Oct 16 '15 at 21:14
  • $\begingroup$ Gaussian mixture modeling works for Gaussian distributions. This is a continuous distribution. $\endgroup$ Oct 16 '15 at 21:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.