I think you shouldn't even delete duplicate entries before feature selection because they're not real dups and data belongs to different patients. You're manually changing the data distribution by doing so. So, let alone after the feature selection, you shouldn't do it even before it.


Conversion rates without sample sizes (i.e. counts of those who convert vs. those who do not) are not useful. 25% could mean 25 out of 100 or 2500 out of 10,000. The precision of the estimate in the latter is greater than that of the former. If your design is a classic two groups-binary outcome, you can use any number of the tests I write about here. ...

Only top voted, non community-wiki answers of a minimum length are eligible