[Retracted] Active Learning for Imbalanced Data: The Difficulty and Proportions of Class Matter

Figure 2

Bias of models trained on class-imbalanced data. (a) Per-class recall and precision on animals_10-LT. (b) Per-class recall and precision on AHE-LT. We train the models for 30 times with different randomly selected training data, and the proportion is the number of each class instance taken in all training data (such that if initially it has 10 instances which belong to the dog class and the whole training data size is 100, then the proportion of the dog class is 10%).

(a)

(b)