3
$\begingroup$

I am working on a case-control study, where I for each individual have high dimensional data (like illustrated in the image).

enter image description here

I would like to do both PCA analysis and Clustering on this data, but it is complicated by the fact that I do not only have several independent columns per individual, but I also have several independent rows.

Do you have any suggestions for what methods can handle this kind of data, or am I forced to summarize the data such that I only have one value per individual per column?

Improve this question
$\endgroup$
2
  • $\begingroup$ It looks like you have 60,000 records. Is this correct or is it potentially a massively larger file? One way to do it would be to use an unsupervised approach rooted in latent class analysis. This could combine PCA with clustering for your mixture of scale types. Apologies for the software-specific recommendation but the developers of Latent Gold software have some excellent papers about as well as solutions to this type of modeling challenge, for instance, here ... statisticalinnovations.com/publicationsarticles/#dfm $\endgroup$
    – user78229
    Commented Nov 13, 2015 at 14:13
  • $\begingroup$ I have 317 tables like the one see above, where has represent cases and half controls. After looking up "Latent class models", it seems that it is what I am looking for. Thanks for you suggestion - it does not matter that it is software-specific, as I have found an R package, that can handle the same kind of analysis too. $\endgroup$
    – Esben Eickhardt
    Commented Nov 16, 2015 at 9:48

1 Answer 1

Reset to default
0
$\begingroup$

http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0007087

I don't know about independent columns but in terms of multiple observations per individual you can use random forests with subject-wise bootstrapping. However you will have to extend it to unsupervised mode.

Improve this answer
$\endgroup$
1
  • 1
    $\begingroup$ Thanks, we have considered looking into random forests as they are good at handling categories and continous variables combindly, but as you say they are often supervised. I will check out the paper. $\endgroup$
    – Esben Eickhardt
    Commented Nov 16, 2015 at 9:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.