Data Science Weekly Newsletter

Data Science Weekly Newsletter

How much should I oversample/undersample?

Instead, you should ask what objectives and constraints you are optimizing for

Data Science Weekly
Sep 12, 2025
∙ Paid
Share

If you've ever had a class imbalance (1:50 or similar) and asked:

"How much should I oversample? 2x, 4x, more?!, Or should I undersample 60%?"

It's the wrong starting point because sampling is a tool, not the goal.

Instead, it's helpful to start by aligning the objective within the constraints you are working with. Think of things like business costs, eval…

This post is for paid subscribers

Already a paid subscriber? Sign in
© 2025 datascienceweekly.org, a service of DATAYOU, LLC
Privacy ∙ Terms ∙ Collection notice
Start writingGet the app
Substack is the home for great culture