Subscribe Bookmark RSS Feed

About finding Confusion Matrix for different Cutoff values in Logistic Regression

sursangeet1

Community Trekker

Joined:

Dec 19, 2012

How do I generate confusion metrics for using different cutoff values? For example, I want to find confusion matrix for cutoff = 0.3 or cutoff = 0.7? etc.

1 ACCEPTED SOLUTION

Accepted Solutions
julian

Staff

Joined:

Jun 25, 2014

Solution

Hi Prakash,

You can generate a new confusion matrix by saving the logistic function, then setting up a column formula, and then using tabulate to make the table. I recorded a quick 2-min video showing the method I would use (it's not terribly complicated, but writing this out in steps started to seem more complicated than it is!).

I hope this helps!

julian

 

edit Sept 19 2017: It looks like the video disappeared some time ago; my apologies! Here is the video again:

 

(view in My Videos)

 

7 REPLIES
julian

Staff

Joined:

Jun 25, 2014

Solution

Hi Prakash,

You can generate a new confusion matrix by saving the logistic function, then setting up a column formula, and then using tabulate to make the table. I recorded a quick 2-min video showing the method I would use (it's not terribly complicated, but writing this out in steps started to seem more complicated than it is!).

I hope this helps!

julian

 

edit Sept 19 2017: It looks like the video disappeared some time ago; my apologies! Here is the video again:

 

(view in My Videos)

 

Craige_Hales

Staff

Joined:

Mar 21, 2013

Nice video!

Craige
AaronNHorvitz

New Contributor

Joined:

Apr 24, 2017

Would you mind sharing a link to the video referenced in the solution?  Thank you! 

 

 

dale_lehman

Community Trekker

Joined:

Jan 29, 2015

I believe this add-in (https://community.jmp.com/t5/JMP-Add-Ins/Alternate-Cut-off-Confusion-Matrix-Add-In/ta-p/22357) does what you are looking for.  Be aware that the overall accuracy of the model is not likely to improve by changing the threshold probability - the main value is that when the costs of misclassification are asymmetric, changing the threshold can reduce the more costly type of errror.

Oleg

Contributor

Joined:

Sep 13, 2017

Hi, would you please indicate where users can find this video?
I can't see any links to it on this Q&A page.

julian

Staff

Joined:

Jun 25, 2014

Hi everyone,

 

I just re-uploaded the video to my original post. My apologies for it disappearing!

-julian

 

 

Oleg

Contributor

Joined:

Sep 13, 2017

Thanks Julian. 

Also, I hope this helps. In JMP Pro, on the model output page, click arrow down in top left corner and select Confusion Matrix. Then open Set Probability Threshold in the appeared new section of the output. This will show a slider you can move to control Recall and Specificity. Unfortunately, it does not seem to calculate Precision and F1 score :( But, I like the dynamic slider for the table :)

 

Screen Shot 09-28-17 at 11.06 AM.PNG Screen Shot 09-28-17 at 11.07 AM.PNG