cluster external validation

Question

I am using ELKI in order to perform location clustering with DBSCAN and OPTICS. My data set include 30 participants but it is not labeled but I do have pair of coordinates (e.g. home, work, etc) as each participant's frequent places.

I want to know that these pair of coordinates belong to which cluster (for each person). One of the way is to check each pair against each of the cluster manually using some minimum distance threshold.

What could be the better way to achieve this?

Actually I want to validate the clusters against the ground truth. — user1124825, Jun 21 '16 at 12:28
The data is not labeled but I do know the frequent locations of people, on average each person has 4 to 6 different locations. — user1124825, Jun 21 '16 at 12:36
but how do I compare that e.g. home corresponds to cluster X? — user1124825, Jun 21 '16 at 12:57
The usual evaluation measures will "align" labels to clusters. But they cannot define "home", you need to provide such labels. — Has QUIT--Anony-Mousse, Jun 21 '16 at 12:59
Would you like to provide pointers about how to provide the labels? — user1124825, Jun 21 '16 at 13:03
Add a column with the label (as *text*) to your data set. ELKI then automatically evaluates the result. — Has QUIT--Anony-Mousse, Jun 21 '16 at 13:04
Let us [continue this discussion in chat](http://chat.stackoverflow.com/rooms/115209/discussion-between-user1124825-and-anony-mousse). — user1124825, Jun 21 '16 at 13:12

score 2 · Answer 1 · answered Jun 28 '16 at 10:52

2

Can you format your input data as this:

123 456 work1
124 457 work1
789 123 home2
123 123 unknown

The labels should be non numeric, that's why I chose "work1", "work2" etc. for this example.

Then ELKI can automatically evaluate the result.

answered Jun 28 '16 at 10:52

Erich Schubert

8,575
2
26
42

cluster external validation

1 Answers1