medium10 pts · first answer countsclass-imbalancedecision-thresholds

Using a model on rare positives

A fraud model outputs probabilities, and only 0.5% of cases are fraud. At the default 0.5 threshold it flags almost nothing. What is the best next step?

A.Pick a threshold from the precision-recall curve based on the cost of each errorB.Delete most of the legitimate transactions from the test setC.Report accuracy, since it is highD.Always use a 0.5 threshold for fairness

Discussion 0

Join free to post in the discussion.

Using a model on rare positives

Discussion 0

Related questions