Define Costs
Use this dialog to specify the cost of errors. For general information about costs,
see Costs.
You specify costs using a cost matrix for Naive Bayes and Decision Tree models. You do not specify costs for a Support Vector Machine model.
Naive Bayes and Decision Tree Models
For NB models, you specify costs using a cost matrix. A default cost matrix, which assumes equal cost for all errors is displayed. The rows of
the cost matrix indicated actual values; the columns, predicted values.
Follow these steps to specify a costs matrix:
- Select the target value.
- Select the kind of erroneous prediction that is most important to avoid:
- False Positives, where a case is incorrectly identified as being the target value
- False Negatives, the default, where a case is incorrectly identified as not being the target value
You can specify costs for both of these options.
- Specify a Weight, that indicates the relative cost of a wrong prediction. For example, if the wrong prediction is two and one-half times as expensive as a correct prediction, specify 2.5 for Weight. The default value for Weight is 1.0, indicating that a wrong prediction has the same expense as a correct one.
- Click Apply to update the matrix with the new weight. You can click Restore to return to the default values.
Support Vector Machine Models
If you are building an SVM model, you specify costs using model weights instead of a cost matrix.
For information about SVM weights, see
SVM Weights.
Finish Cost Definition
When you have finished specifying costs, click OK to return to the calling page.
Copyright © 2006, 2008, Oracle. All rights
reserved.