I am dealing with a situation where I need to use a joint matrix of approximately gaussian distributed data (limits - Inf to + Inf) and beta-distributed data (both produced using different experimental methods for the same samples) to generate clusters. So far I have tried clustering a Random Forest proximity matrix , and the thing that works best is an unconventional approach where I do not model the proximity matrix explicitly as a dissimilarity.

I am looking for advice on whether transforming everything to approximately Gaussian using either Z-scores or a logit transform of the beta-distributed variables is a good way of creating a matrix suitable for direct clustering.

  • 1
    $\begingroup$ Unclear, for me. Are you going to do cluster analysis or to generate clusters? $\endgroup$
    – ttnphns
    Jun 12 '16 at 7:41
  • $\begingroup$ If the attributes are approximately Gaussian / approximately beta distributed, then there shouldn't be any clusters there! $\endgroup$ Jun 12 '16 at 9:48
  • $\begingroup$ A) I am looking to generate clusters B) There are 1259 features across 145 samples - each one of them is approximately Gaussian or beta distributed - but there are clear clustering patterns visible in each of the datasets - a heatmap with Hierarchical clustering dendrograms confirms this. $\endgroup$ Jun 12 '16 at 13:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.