Receiver Operating characteristic
Graphical representation of the performance of a binary classifier at different classification thresholds; for Binary Classification but stochastic model not continuous model by changing threshold.
- X - False Positive Rate (FPR) Fall-out
- Y - True Positive Rate (TPR) Sensitivity
Random classifier is linear line
Perfect model

Fine model

Inverse Perfect Model

Bad Model


ROC AUC
Area Under the Curve
- Ideal model, ROC AUC=1
Youden's Index
0, 1에서 가장 가까운 점 찾기
