0
$\begingroup$

I am dealing with a classification task of a binary target variable (company failure prediction yes or no) for a university project. I was wondering, should the standardisation of numerical variables be carried out before or after the rebalancing technique of the target variable?

Improve this question
$\endgroup$
7
  • $\begingroup$ I assume that by rebalancing you’re referring to some kind of matching procedure - so it depends on how you’re carrying out the matching; some methods will be invariant, others won’t $\endgroup$
    – Randy Savage
    Commented May 14 at 19:03
  • $\begingroup$ I used the SMOTE technique $\endgroup$
    – Elena
    Commented May 14 at 19:08
  • 1
    $\begingroup$ Welcome to Cross Validated! 1) Why did you run the SMOTE? Oversampling is typically a solution to a non-problem, and SMOTE does not even seem to be particularly good at synthesizing new points. 2) Why do you want to standardize? It's not a theorem that you have to standardize, even though it can make sense in some situations. $\endgroup$
    – Dave
    Commented May 14 at 19:37
  • $\begingroup$ I performed the standardisation because I have x1-x15 variables (two categorical ones that I turned into dummies) whose scale and meaning I do not know. As for SMOTE, looking at similar tasks on github and kaggle it was used to balance the minority class of the target. I don't know if that makes sense $\endgroup$
    – Elena
    Commented May 14 at 20:01
  • 1
    $\begingroup$ I performed the standardisation because I have x1-x15 variables (two categorical ones that I turned into dummies) whose scale and meaning I do not know. This does not seem like a reason to standardize. If you are doing deep learning and want your values to be in $[0, 1]$ for computational reasons, some kind of standardization could make sense. If you want to interpret your coefficients in terms of how many standard deviations the variable must change to elicit a certain change in the outcome, standardization could make sense. My answer to your question would be different for each situation. $\endgroup$
    – Dave
    Commented May 14 at 20:05

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.