False positives and negatives in a fraud model

Question

A model flags transactions as "fraud" or "not fraud." Treat the null $H_0$ as "this transaction is legitimate."

Connect type I and type II errors to the classifier's confusion matrix, relate them to precision and recall, and explain which error a fraud team usually prioritizes.

Accepted Answer

The confusion matrix. With H0 = "legitimate," flagging a transaction as fraud is rejecting the null: - Type I error (false positive): flag a legitimate transaction as fraud. You reject H0 when it is true. - Type II error (false negative): clear a fraudulent transaction. You fail to reject H0 when it is false. Laid out as a 2 	imes 2 table of (truth) versus (model call), the four cells are true positives (fraud caught), false positives (type I), true negatives (legit cleared), and false negatives (type II, fraud missed). Precision and recall. - Recall (sensitivity, 1 - \beta, the model's powe…

False positives and negatives in a fraud model

Your answer

More Statistics questions