1
$\begingroup$

I have a dataset with class imbalance about (1:10). I applied a SMOTE method by slitting the dataset into training and dev.

I over sampled the training set using a SMOTE method and divided this set into training and test to build a RF classifier.

When I evaluate the RF classifier on the dev set which has not be over sampled it preforms quite poorly.

Is it standard that I would apply SMOTE to the dev set as well?

$\endgroup$
0
$\begingroup$

See the advice on this topic here: https://beckernick.github.io/oversampling-modeling/

$\endgroup$
0
$\begingroup$

I think you are giving your subsets bad names, look here fer reference.

By the way, NO, of course not, you should only apply SMOTE on training data, your goal is to train a model that predicts well real data, not synthetic ones, SMOTE is just an escamotage for achieving better training.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.