0
$\begingroup$

I have a total of 32 observation that I would like to cluster. The data avaiolable comes in 2 formats:

  • Continuous data
  • Summarized categorical data

What I mean with summarized categorical data is that I originally had categorical (e.g. band 1, 2 or 3) data which I summarised (band 1 = 50%, band 2 = 30% and band 3 = 20%) for each observation.

What are some common clustering algorithms that I can use with this mix of data? What do I need to be careful of when mixing these two types of variables in one model?

Thanks!

$\endgroup$
0
$\begingroup$

Generally clustering is very sensitive to scales, so rule of thumb is to transform all the variable to comparable scales wherever possible. In your case all the variables are already in fact continuous, since your summaries are %%.

I would start with - (1) checking the distributions for all variables, (2) if they generally fit Gaussian profiles, computing z-scores (3) going with standard squared euclidean distance / hierarchical clustering

Since you have only 32 objects, hierarchical clustering should serve you best - (1) it's visual, (2) does not require initial assumptions on cluster number and cluster centres (3) and not sensitive to order of cases, which is important with small samples

$\endgroup$

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.