How do you adjust for oversampling?

Curious

After training the model?

Telcontar120

If you have sampled either way (up or down) and have not used weights, you will probably need to go back to your original dataset (or a non-modified sample thereof) and recalibrate your scores.  The score rank ordering should be preserved but the absolute relationship between scores and probabilities will likely need adjustment.

rfuentealba

Hi @Curious,

If you have two classes that aren't well balanced, you can use the Sample operator to balance the data to downsample the bigger class. However, to oversample the smaller class you can use the Sample - Balance operator in the Mannheim RapidMiner Toolbox extension.

All the best,

Rodrigo.

varunm1

Hi @Curious

Can you elaborate your question? If you are asking about sampling (up or down), then @rfuentealba mentioned in his comment, but I am not sure about "After training the model?" statement in your question.