What is the primary risk of using only accuracy as the evaluation metric for a factory defect detection model when defects are rare?
-
A
Accuracy cannot be calculated for imbalanced datasets
-
B
A model predicting 'no defect' always will achieve high accuracy but miss all defects
-
C
Accuracy overestimates false positives in rare-event scenarios
-
D
Accuracy is undefined when classes are unbalanced