Timeline for Summarizing multiple clustering results
Current License: CC BY-SA 3.0
10 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jan 5, 2016 at 17:20 | comment | added | user32398 | You could just run ANOVA, and then use a modified Type I error level (Bonferroni or Sidak $\alpha^* = \alpha /\#tests$) as the p-value criterion, or use Benjamini-Hochberg FDR. From a machine learning perspective, straightforward inferential hypothesis testing won't cut it alone -- you'd have to use CV at every step. | |
| Jan 5, 2016 at 16:02 | comment | added | Ellis Valentiner | @LEP I expect that the clusters within each group will differ – so I expect that the ANOVA will nearly always be "significant". | |
| Jan 5, 2016 at 15:40 | comment | added | user32398 | Regarding CV during ANOVA, if you randomly generated 100 values of a variate in 3 groups, and ran ANOVA to determine if the three averages (based on $n=100$ in each group) are significantly different you can obtain significance by chance alone. However, if you e.g. sampled 20%,30%,40%, or 50% of the objects from each group and ran ANOVA on the smaller sample sizes, then kept a running average of -log(p-value) for every fold and repeated this, say, 10 times, you could minimize chances of a obtaining significant results by chance alone. | |
| Jan 5, 2016 at 15:35 | comment | added | user32398 | Sure, one-way ANOVA will work, but recall that will only address one feature across all the groups (univariate). You could test many of the same features across many groups using a Hotelling test (MANOVA). Also, through chance alone, you could get significantly different means, so you'd need to partition to objects in each group into folds (cross-validation) and then keep an average value of e.g. -log(p-value). | |
| Jan 5, 2016 at 15:19 | comment | added | Ellis Valentiner | @LEP I am thinking that the clustering aspect is less important and the comparison of the groups is more so. Can I just do a one-way anova for each group? | |
| Jan 5, 2016 at 14:24 | comment | added | user32398 | After clustering within a group, you could calculate several cluster quality indicators (silhouette index, Dunn's index, Davies-Bouldin index, Hubert's $\Gamma$), and then compare these across groups. | |
| Jan 5, 2016 at 13:41 | comment | added | Ellis Valentiner | @DavidG.Stork the purpose is to reduce the information in each "group" by "prototypes" (i.e. clusters) and to then compare the groups in terms of their prototypes. | |
| Jan 5, 2016 at 10:10 | history | edited | Has QUIT--Anony-Mousse | CC BY-SA 3.0 |
edited title
|
| Jan 5, 2016 at 1:00 | comment | added | David G. Stork | What is the purpose of your analysis? Are you trying to find the group with the greatest inter-group variation? | |
| Jan 4, 2016 at 23:05 | history | asked | Ellis Valentiner | CC BY-SA 3.0 |