Skip to main content
D
DF_100273561-Visitor
1-Visitor
February 7, 2022
Solved

Question regarding AUC Values

  • February 7, 2022
  • 1 reply
  • 1743 views

Hello all,

 

I have created a time series model with a boolish goal variable. I set the validation hold out to 20 %. But now I get Area under the curve (AUC) values below 0,5. Sometimes even zero. I always thought that the values of the AUC range from 0,5 to 1?

 

Am I missunderstanding something? Or what does a AUC value below 0,5 mean?

 

Thank you already in advance.

 

Best regards

David

Best answer by nsampat

@DF_10027356 ,

 

Thank you for posting your question to the PTC Community.

 

Something to understand, you can have a ROC value below 0.5 as 0.5 only indicates that there is no class separation capacity. A value of 0 indicates that the value is reciprocating to the class.

 

I found this blog that may help provide additional details on your question: External Link - Understanding AUC - ROC Curve

 

Please let me know if you have any additional questions.

 

Regards,

 

Neel

 

1 reply

N
nsampat17-PeridotAnswer
17-Peridot
February 7, 2022

@DF_10027356 ,

 

Thank you for posting your question to the PTC Community.

 

Something to understand, you can have a ROC value below 0.5 as 0.5 only indicates that there is no class separation capacity. A value of 0 indicates that the value is reciprocating to the class.

 

I found this blog that may help provide additional details on your question: External Link - Understanding AUC - ROC Curve

 

Please let me know if you have any additional questions.

 

Regards,

 

Neel

 

    D
    DF_100273561-VisitorAuthor
    1-Visitor
    February 7, 2022

    Hello Neel,

     

    thank you very much for your answer. I understand the problem now. So maybe I can increase my AUC when I just exchange the classification in my training data set (a zero becomes a one and a one becomes a zero)? Or is this problem maybe also caused by the algorithms I am using (Neural Net, Gradient Boost, Decision Tree and Logistic Regression)?

     

    Thank you for your help.

     

    Best regards 

    David

    N
    nsampat17-Peridot
    17-Peridot
    February 7, 2022

    @DF_10027356 ,

     

    Much of what you are asking is dataset specific, and I am unable to provide any advice on how to complete your solution.

     

    If you need assistance with use case creation and interpretation, I can refer you to the ThingWorx Analytics Center of Excellence consultants.

     

    Regards,

     

    Neel

     

    Terms & ConditionsAccessibility statement