0

As it is written here, you should standardize the data before applying SMOTE. If I inverse the standardscaler action with inverse_transform after using SMOTE, will it know how to inverse the synthesized observation?

Amit S
  • 27
  • 7
  • "Correct" for what purpose? – whuber Apr 19 '21 at 17:09
  • 2
    Frank Harrell, a member on here and the founding chairman of the Biostat department at Vanderbilt, suggests not using SMOTE at all: https://twitter.com/f2harrell/status/1062424969366462473?lang=en – Dave Apr 19 '21 at 17:19
  • The SMOTE synthesizes new observations. Does the inverse_transform know how to inverse them? – Amit S Apr 19 '21 at 17:19
  • 1
    According to https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.StandardScaler.html, the scaling is simply a change of units of measurement. I can't speak to what the software might "know," but changing the units back is simple. – whuber Apr 19 '21 at 17:35
  • But the SMOTE creates a new data point from the original point that was standardized. Could the inverse_transform command succeed inversing back a point that wasn't standardized by the StandardScaler? – Amit S Apr 19 '21 at 18:16

0 Answers0