The Ensemble node combines two or more model nuggets to obtain more accurate predictions than can be gained from any of the individual models. By combining predictions from multiple models, limitations in individual models may be avoided, resulting in a higher overall accuracy. Models combined in this manner typically perform at least as well as the best of the individual models and often better.
This combining of nodes happens automatically in the Auto Classifier and Auto Numeric automated modeling nodes.
After using an Ensemble node, you can use an Analysis node or Evaluation node to compare the accuracy of the combined results with each of the input models. To do this, make sure the Filter out fields generated by ensembled models option is not selected in the Ensemble node settings.
Each Ensemble node generates a field containing the combined scores. The name
is based on the specified target field and prefixed with
$XR_, depending on the field measurement level: flag,
nominal (set), or continuous (range), respectively. For example, if the target is a flag field named
response, the output field would be
Confidence or propensity fields. For flag and nominal fields, additional confidence or propensity fields are created based on the ensemble method, as detailed in the following table.
|Ensemble method||Field name|
Highest confidence wins
|Average raw propensity||
|Average adjusted raw propensity||