turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

- JMP User Community
- :
- Discussions
- :
- Discussions
- :
- ROC curve on Neural platform

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Sep 12, 2017 11:55 PM
(1135 views)

Hello everybody,

I am working with a binomial variable and I am testing different analyses.

When I do a logistic regression and display the ROC curve, only one curve is displayed.

However if I display this report from the Neural platform, 2 curves are displayed and I don't understand why.

From what I understand the creation of a ROC curve is based on the probabilities associated with each observation (no matter how they are calculated). And I thought that with two levels we always have only one curve...

Could you provide me some information about this point please?

Solved! Go to Solution.

1 ACCEPTED SOLUTION

Accepted Solutions

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Sep 13, 2017 8:28 AM
(2079 views)

When you are fitting a logistic regression model, you are asked to choose which level of your binomial variable is the "positive" level. That results in only a single ROC curve being needed -- you know which level is positive.

When you are fitting a neural network model, you are NOT asked which level is positive. Therefore, the neural network platform will generate two curves: one for each possible result being the "positive" value. If you look at the areas under the ROC curve, they should be identical. The only difference is which level of your response is positive.

Dan Obermiller

2 REPLIES

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Sep 13, 2017 8:28 AM
(2080 views)

When you are fitting a logistic regression model, you are asked to choose which level of your binomial variable is the "positive" level. That results in only a single ROC curve being needed -- you know which level is positive.

When you are fitting a neural network model, you are NOT asked which level is positive. Therefore, the neural network platform will generate two curves: one for each possible result being the "positive" value. If you look at the areas under the ROC curve, they should be identical. The only difference is which level of your response is positive.

Dan Obermiller

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

Sep 13, 2017 8:37 AM
(1116 views)

Thank you Dan that's clearer now!