Das Szenario
A fraud analytics team needs more labeled fraud cases than the real dataset provides, so it uses a GAN to synthesize realistic fraud transactions and then trains a classifier on the augmented data.
Was diese Zeichnung zeigt
Die dahinterstehenden Entscheidungen verstehen.
01
Whether synthetic quality is sufficient before combining data
02
Whether final classifier performance on real holdout is acceptable for deployment
Reuse this pipeline when real fraud labels are scarce and a GAN can generate plausible synthetic examples for model training.