Overview
Uncertainty sampling refers to the process of finding records in a dataset where a model is most uncertain of the true label. This can useful to understand the situations where the model is simply guessing at the right answer. This servers both an evaluative function, but is also used to indentify records that the model needs to spend more resources training on, and may suggest ways to edit the model in order to make it more effective.
The challenge with uncertainty sampling is coming up with a measure of uncertainty for model being used.