I woul like to stratify a multivariate dataset in different stratas in such a way that the elements of each strata are similar. My idea is to have different groups whith the element within a group close in term of distance. My idea is to divide a multivariate sample in a similar way that univariate date is separated into deciles, quintiles, quartiles, etc.

I've been thinking on using a k-means algorithm but this algorithm does not separate groups in similar size.

Do you know of a standard procedure to obtain my objective?

My dataset has nearly one milion rows, I have been not able to apply a hierarchical approach.

  • $\begingroup$ Calculate a Mahalanobis distance and split it into percentiles to get groups of equal size. $\endgroup$ – AdamO Mar 2 '19 at 14:40
  • $\begingroup$ But in this way we can have elements in a group which are far form others elements in the same group (I am thinking on antipodal elements). $\endgroup$ – marc1s Mar 2 '19 at 15:27

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.