Two observations have the same imputed value but only one was originally missing. How can the model distinguish them?
Classic MLLevel: IntermediateMissingness indicators
Two observations have the same imputed value but only one was originally missing. How can the model distinguish them?
Hints
- Add a separate binary feature.
- The imputed number alone loses provenance.
Answer
Append missingness indicators, for example add_indicator=True. This preserves the fact of missingness, which may carry signal, without pretending the imputed value was measured.
Source and licence: scikit-learn 1.7.2 — impute · BSD-3-Clause · answer: Quantoros editorial