Comparing AUC vs accuracy

Asked Jun 11 '14 at 01:05

Active Jun 11 '14 at 11:14

Viewed 161 times

I understand this question has been asked many times however, i am unable to understand the answers well enough and apply to my situation.

I have attached 2 screenshots of my model. There are 5 class labels

Firstly this is a model with a lower accuracy but with a higher ROC area.

enter image description here

Then i have a model with a higher accuracy but a lower ROC area.

enter image description here

How do i decide which model is a better one if the cost for misclassification for all classes is the same

edited Jun 11 '14 at 11:14

asked Jun 11 '14 at 01:05

aceminer

How can a model have multiple ROC Areas? This makes no sense to me. – Calimo Jun 11 '14 at 06:52
@calimo: I'd guess there are five classes & it's one of those extensions of ROC to the multinomial case. Perhaps aceminer can clarify. – Scortchi - Reinstate Monica Jun 11 '14 at 08:15
@Calimo Yes there are 5 class labels – aceminer Jun 11 '14 at 11:14

0 Answers0