Timeline for K Means giving poor results
Current License: CC BY-SA 3.0
9 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Mar 17, 2016 at 22:55 | comment | added | Has QUIT--Anony-Mousse | No, k-means is not assuming Gaussian mixtures. GMM is the usual EM approach; and it is just as much affected by skewed distributions, because it assumes Gaussians which are non-skewed. Apply exp() to all values, and your clustering result will become substantially worse. | |
| Mar 17, 2016 at 20:41 | comment | added | Tu N. | @Anony-Mousse I don't think your claim is correct. K-means can be seen as a EM algorithm on Gaussian Mixtures, so skewedness doesn't make K-means work better or worse. Skewedness of data is likely to skew up the initialization process of centroids, which make K-means trap in local optima. That's being said, even if you have balanced data, K-means might not work because of bad initialization. | |
| Mar 17, 2016 at 6:56 | comment | added | Has QUIT--Anony-Mousse | For 1 dimensional and 2 dimensional data, I suggest you also visualize the data (and any result) and discuss what is interesting, bad, difficult, desired on the plot. Datascience is about telling a story, and images help a lot there. | |
| Mar 17, 2016 at 6:54 | comment | added | Has QUIT--Anony-Mousse | @Sreejithc321 I don't know if you have a "typical clusteeing problem" because you have not stated the problem. What are valid answers to your problem, and what makes one answer better than another. Until you specify this, random assignment is a valid solution to your problem. | |
| Mar 17, 2016 at 6:51 | comment | added | Has QUIT--Anony-Mousse | @TuN. I already gave an example... beware, there are two kinds of skewedness. Some are fine, e.g. if your data is 1000 objects from N(1, 1) and 100 from N(10,1) then this is will be considered a skewed variable, but probably fine for k-means because the clusters are well separated and not skewed themselves. | |
| Mar 17, 2016 at 6:36 | comment | added | Sreejithc321 | yes I got your point, so this is not a typical clustering problem. right ? Then the best way is to sort salaries based on score and divide them to 'K' categories right ? Also is there a way to calculate the best number for 'K' ? suggestions please. | |
| Mar 17, 2016 at 6:34 | vote | accept | Sreejithc321 | ||
| Mar 17, 2016 at 3:31 | comment | added | Tu N. | do you mind explaining what's "translation invariant"? Do you have proof or any reference in literature to claim that k-means does not work on skewed data? | |
| Mar 16, 2016 at 19:56 | history | answered | Has QUIT--Anony-Mousse | CC BY-SA 3.0 |