Explaination drawn from this youtube video -
youtube source
###True Positive
when the actual classification is positive, how often does the classifier reports positive
$dfrac{true space positive}{all space positives}$
###False Positive
when the actual classification is negative, how often does the classifier incorrectly predict positive
$dfrac{flase space positive}{all space negatives}$
ROC - Receiver operating characteristic -
A graph that show true positive rate
against false positive rate
at different threshold setting. It can be used to select possibly optimal models and discard the suboptimal ones. ROC analysis is related in a direct and natural way to cost/benefit analysis of diagnostic decision making.
AUC - Area under curve
Area under the ROC curve; how well the classifier separate the ditribution
0.5 => random guessing // 1.0 => perfect
Some properties
- AUC is useful when the distribution is highly unbalanced
- ROC can be extended to multiple classifiers
- How to set the threshold is a business decision, the tradeoff between
MAXIMIZING true positive
vsMINIMIZING false positive
近期评论