1
$\begingroup$

I have a D-dimensional dataset composed of exactly two clusters (this is known) for which I have no labels; the clusters can potentially be wildly imbalanced.

I'm after a soft (or fuzzy) clustering method to assign probabilities to each element of belonging to either cluster. So far I've been able to come up with basically two:

Then there are methods that apply hard clustering that could perhaps be softened by re-running varying the inputs (and averaging all the iterations?):

And finally there's also those methods that I'm not sure whether they can be applied in an unsupervised way at all:

Am I missing some method? Did I miss-classify any of the above? Is any method more suited to my particular issue than the rest?

Any insight will be much appreciatted.

$\endgroup$

1 Answer 1

0
$\begingroup$

UPDATE

(Looking back on the OP, I would recommend running FKM and GMM, then try to publish the analysis based on that, or give a talk based on use of FKM/GMM. There is nothing wrong with use of FKM/GMM for your OP. You may be "missing" a lot other unknown/unpopular methods which develop probabilities -- so construction of list that misses nothing would become open-ended).

Only fuzzy k-means provides "membership function" values for each object, and GMM provides cluster-specific probabilities, $P(cluster|x)$. "Crisp" k-means cluster analysis (opposite of fuzzy, i.e., the typical k-means) can provide something like a cluster-specific probability if you assume that Euclidean distance to the closest centroid represents a probability. There may be more methods that do this, but generally speaking, most other unsupervised methods are based on distance metrics, not probabilities.

I wouldn't get "hung up," i.e., interested on being exact about all the various unsupervised methods and which ones do and don't provide cluster-specific probabilities. If you are writing a review, book chapter, or a report for school, you shouldn't ask this forum about it.

Have you grasped the full literature on this topic? You'd get a much better answer doing your own research.

$\endgroup$
2
  • $\begingroup$ see the updated answer - if a final list were required, it would be an open-ended question. $\endgroup$
    – user32398
    Mar 15, 2020 at 18:38
  • $\begingroup$ Also, did you run FKM and GMM yet, and submit the paper or give the talk? What did the reviewers (audience) say? $\endgroup$
    – user32398
    Mar 15, 2020 at 18:42

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.