0
$\begingroup$

As per the references 1. http://www.sthda.com/english/wiki/assessing-clustering-tendency-a-vital-issue-unsupervised-machine-learning 2. http://www.listendata.com/2016/01/cluster-analysis-with-r.html

The Hopkins statistics value close to '0' is also clusterable. Is this correct? Or the value should always be close to '1' for clusterable?

$\endgroup$
1
$\begingroup$

I don't think the Hopkins statistic is useful for this purpose at all.

It is essentially a test for a uniform distribution.

But not having a uniform distribution does not mean the data is suitable for cluster analysis.

For example a single Gaussian distribution (unimodal!) will score high on this test, but the data doesn't have multiple clusters but all points are from the same distribution.

| improve this answer | |
$\endgroup$
  • $\begingroup$ Thanks, Anony for your answer. Can you point out any standard technique which test the data to tell whether data is suitable for clustering or not? My data is a multi-dimensional data. $\endgroup$ – Justin Martin Oct 4 '16 at 12:05
  • $\begingroup$ I don't know any. Run clustering and inspect the results. The hardness of clustering is that there is no objective truth. $\endgroup$ – Has QUIT--Anony-Mousse Oct 5 '16 at 6:01
1
$\begingroup$

One way to look for number of clusters is through VAT (Visual Assessment of Tendency), the reference you have provided: Alboukadel Kassambara also talks about that. This is a more frequently used measure by the machine learning community.

You can also explore Hierarchical tree and look for patterns of your interest.

At the end, it is subjective and application dependent.

| improve this answer | |
$\endgroup$
1
$\begingroup$

The book "Data mining concepts and techniques"(by Jiawei Han) in the chapter of clustering talks about Hopkins Statistic(value range=(0,1)):

(1) high score-->uniform distribution-->no cluster

(2)low score-->not uniform distribution-/->(may be not)cluster.

| improve this answer | |
$\endgroup$

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.