- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Lift curve allows modelers to compare models. But this is done visually. Can lift chart be translated into a single metric so that comparison can be made even easier?
Here is a lift curve example online:
http://www.kdnuggets.com/gpspubs/kdd99-est-ben-lift/sld022.htm
Accepted Solutions
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Hi, Aha. Do you really want to compare entire curves?
Data miners usually compare lift curves at a particular sample depth. In Enterprise Miner, the Model Comparison node uses a default depth of 10 when Lift is the model selection criterion, but you can change the default depth.
Ray
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I'm no expert in this area but I think you are after the Gini index or Somer's D or calculating the area under a ROC curve. This link contains references to all of these:
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I forgot to mention this is a question for regression problem. As far as I know, ROC is for classifier problems.
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
If its a binary outcome then PROC LOGISTIC is mentioned as well in the link page.
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Hi, Aha. Do you really want to compare entire curves?
Data miners usually compare lift curves at a particular sample depth. In Enterprise Miner, the Model Comparison node uses a default depth of 10 when Lift is the model selection criterion, but you can change the default depth.
Ray