2
$\begingroup$

If the ground truth of the class/cluster/segment that our observations belong to, is known in advance, why would someone choose to perform clustering instead of classification? In fact, doesn't the problem "automatically" become a classification problem?

This question came to my mind as I was going through some clustering performance evaluation criteria. I came across the Rand index, an evaluation metric that requires the ground truth to be known in advance, hence my question.

$\endgroup$
5
  • $\begingroup$ "doesn't the problem "automatically" become a classification problem?" What problem? You sound like there was a problem first (which, upon obtaining a ground truth variable, "becomes" a classification kind), so what was that problem? $\endgroup$
    – ttnphns
    May 11 at 7:43
  • $\begingroup$ The problem that we are trying to solve using clustering. Say we want to surface clusters of observations so that observations in the same group are as similar as possible and observations in different groups are as dissimilar as possible. What I'm trying to say is that if I had the ground truth of these observations in the first place, I wouldn't use them to evaluate my clustering but to perform classification instead $\endgroup$ May 11 at 11:15
  • 2
    $\begingroup$ These are two unrelated tasks, classification and clustering. I may know the ground truth classes, still may want to partition objects into groups as much similar as those classes as possible, but without using the truth knowledge during the process of working out the rules for partition. It is like creatively learning without a tutor, then meeting a tutor later at exams. $\endgroup$
    – ttnphns
    May 11 at 12:49
  • $\begingroup$ FYI, you might want to read about the most important external cluster validity criteria, their formulas, in the "Compare partitions" collection on my web-page. $\endgroup$
    – ttnphns
    May 11 at 13:00
  • $\begingroup$ I think the answer is obvious when you consider any type of prediction problem in general. "Why would I want to try to predict the stock market based on indicators, why would I try to predict cancer diagnoses based on test results" the answer is you want to build a model to apply to data where you don't know the truth (eg predict future stock price, predict diagnosis given only test result, etc). $\endgroup$
    – bdeonovic
    May 11 at 18:35

1 Answer 1

3
$\begingroup$

You would want to cluster instead of classifying when the real-world problems don't share the same categories as the evaluation set you use.

For instance, let's say you know the true clusters of a small network into six groups. If you were to learn a classifier, then for all future networks you'd only be able to split them into six groups. By contrast, with clustering, you can divide them into arbitrary numbers of groups, which may be more appropriate.

You validate the clustering model on datasets that you know, in the hopes that it generalizes better to ones that you don't.


At a broader level, if the ground truth is known, then there's nothing left for you to predict—so the only reasonable goal is trying to understand (or explain) that structure. You can have competing hypotheses (competing models) of how that ground truth structure arose. Some might be drawn from clustering literature; others from classification literature. Each has its own set of inductive biases.

$\endgroup$
6
  • $\begingroup$ With respect, I’ve answered the question you asked. This website is set up so that if you have another question, you can post it as a separate question. If you choose to take advantage of the opportunity, it would be helpful to clarify what you mean by “other ones”. $\endgroup$ May 11 at 12:09
  • $\begingroup$ You did answer the question I've asked indeed but I want to understand if this answer is complete. Meaning that there's NO other case when you'd use clustering instead of classification (given that you have the ground truth) other than the one you mentioned. I hope that makes sense. I'm not challenging the validity of the answer I'm just wondering if there's more $\endgroup$ May 11 at 13:57
  • $\begingroup$ Broadly speaking, clustering is used to discover latent structure. Classification is used to link to a known structure. It’s hard to get more broad than that, so there’s not much left to cover. I hoped that the answer I originally provided, by being concrete, would make it more accessible. $\endgroup$ May 11 at 14:17
  • $\begingroup$ The question is not about when would one use clustering in general. The question is asking when would one use clustering given that the structure (ground truth) is already known. If this is unclear from my description/title please feel free to edit it. $\endgroup$ May 11 at 15:06
  • 1
    $\begingroup$ With respect, it's your question and therefore your responsibility to make clear, not mine. If the ground truth is known, then there's nothing for you to predict—so the only value is in trying to understand (or explain) that structure. You can have competing hypotheses (competing models) of how that ground truth structure arose. Some might be drawn from clustering literature; others from classification literature. Each has its own set of inductive biases. $\endgroup$ May 11 at 15:10

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.