Should you rarefy for q2-sample-classifier?


Is it best practice to use a rarified table for supervised machine learning or not?

Thank you,


Hi @Zach_Burcham,
Great question! In part this boils down to what type of estimator you are using — some are really sensitive to normalization/scaling issues, while others (like random forests) are quite robust.

So if you are using random forests rarefying probably will not change your results (this has been my experience and what I’ve seen reported in the literature); but rarefying won’t hurt either and either rarefying or performing some other type of normalization prior to using these methods is probably just good practice, as has been recommended elsewhere in the literature.


This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.