-2
$\begingroup$

I have a dataset with 100 nominal categorical variables with two levels each, for example:

Do you smoke? Yes/No

Do you like to dance? Yes/No

etc.

How can I cluster this dataset for try to group/create "profiles"?

I know a few type of clustering algoritms, like PCA, K-neighbors etc. But I think that I can't just use this type of algoritms with non-quantitative variables, is it?

Thanks you.

$\endgroup$
  • $\begingroup$ Did you search the site? There are so many posts about clustering categotical data. $\endgroup$ – ttnphns Jan 5 at 14:15
0
$\begingroup$

You can do two data pre-processing transformations: mapping non-numeric data into binary dummies is a method. Check this article'sdata section. Your data is basically similar. After data pre-processing , SVM or another data classifier methods can be applied.

$\endgroup$
0
$\begingroup$

For this problem I recommend using latent class clustering. This kind of clustering algorithm is appropriate for categorical responses. Your assessment that methods such as k-means clustering are probably not a good choice is correct.

Latent class clustering is a model-driven clustering algorithm. It assumes that we observe a mixture of a finite number of "types". Each type is defined by a probability distribution over the vector of responses. In your case, a type would be defined as a vector of 100 probabilities, where the jth probability gives the probability of selecting "yes" on the jth question.

Latent class clustering is implemented as an R package.

$\endgroup$

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.