Gains-Based Selection

The Gains-Based Selection dialog box enables you to automatically select terminal nodes with the best (or worst) gains based on a specified rule or threshold. You can then generate a Select node based on the selection.

1. On the Gains tab, select the node-by-node or cumulative view and select the target category on which you want to base the selection. (Selections are based on the current table display and are not available for quantiles.)
2. On the Gains tab, from the menus choose:

   Edit > Select Terminal Nodes > Gains-Based Selection

   Select only. You can select matching nodes or nonmatching nodes—for example, to select all but the top 100 records.

   Match by gains information. Matches nodes based on gain statistics for the current target category, including:

   • Nodes where the gain, response, or lift (index) matches a specified threshold—for example, response greater than or equal to 50%.
   • The top n nodes based on the gain for the target category.
   • The top nodes up to a specified number of records.
   • The top nodes up to a specified percentage of training data.
3. Click OK to update the selection on the Viewer tab.
4. To create a new Select node based on the current selection on the Viewer tab, choose Select Node from the Generate menu. See the topic Generating Filter and Select Nodes for more information.

Note: Since you are actually selecting nodes rather than records or percentages, a perfect match with the selection criterion may not always be achieved. The system selects complete nodes up to the specified level. For example, if you select the top 12 cases and you have 10 in the first node and two in the second node, only the first node will be selected.