#f1-score

Question 1: Select all that apply

You evaluate a fraud classifier with threshold-based metrics and try to fix its class imbalance. Which statements are true? Select all that apply.

The average precision (AP) score at threshold 0.5 is the same as the F1 score1
class_weight="balanced" is likely to decrease false negatives but increase false positives2
Along an ROC curve, the threshold goes from 1 at the lower left to 0 at the upper right3
A macro-averaged recall gives every class the same weight, whatever its size4