I am just wondering about this issue brought up by our teacher about a drawback of K-means being unable to detect small clusters. It's homework that we should come up with ideas about why this is so and propose a solution to address it.

I know that K-means has an assumption that each cluster should have roughly the same number of observations, so perhaps if say 3 clusters have ~1000 observations, the 4th one "shouldn't" have 10 observations? Am I right in saying that?

I read Clustering: k-means alternatives when its assumptions do not hold and Is it true that K-Means has an assumption "each cluster has a roughly equal number of observations"? and lastly this How to understand the drawbacks of K-means.

Is it something to do with minimizing the SSE that causes this issue? Any insights are appreciated.

  • $\begingroup$ Not sure, but if you know the exact number of clusters and its centroid then chances are you will get your true clusters. But in case of small clusters(4th cluster you mentioned), the cluster might have more than the actual instances assigned to it which may belong to other clusters since instances belonging to other clusters might actually be nearer to the centroid of the 4th cluster. But it all depends on how the data is distributed in the space. $\endgroup$
    – Jerry
    Oct 16, 2019 at 17:25


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Browse other questions tagged or ask your own question.