#imbalanced-data

Question 1: Single choice

You are planning a model that flags defective parts from photos. Defects are rare, and every flagged part is re-checked by a worker. How should the model's success be measured?

Recall on defective parts, together with how many parts get flagged for re-checking1
Overall accuracy, because it sums up every prediction in a single easy-to-read number2
The training loss that the model reaches at the end of its final training epoch3
Precision alone, the share of flagged parts that are truly defective4

Timed test