Glossary term
Glossary term
Foundations
The process of replacing a missing value with an acceptable substitute. When a value is missing, you can either discard the entire example or you can use value imputation to salvage the example.
For example, consider a dataset containing a temperature feature that is supposed to be recorded every hour. However, the temperature reading was unavailable for a particular hour. Here is a section of the dataset:
A system could either delete the missing example or impute the missing temperature as 12, 16, 18, or 20, depending on the imputation algorithm.
For example, consider a dataset containing a temperature feature that is supposed to be recorded every hour. However, the temperature reading was unavailable for a particular hour. Here is a section of the dataset:
A system could either delete the missing example or impute the missing temperature as 12, 16, 18, or 20, depending on the imputation algorithm.
Created for this library
A retail analytics team uses value imputation on missing customer ages with the regional median before training a propensity model.
A medical team uses model-based value imputation for missing lab values so the downstream risk model can run on every patient.
A risk modeling team uses value imputation with indicator flags so the production model knows which inputs were originally missing.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License