Impute Missing Values

For decision trees, missing values are not problematic. Surrogate splitting rules enable you to use the values of other input variables to perform a split for observations with missing values. In SAS Enterprise Miner, however, models such as regressions and neural networks ignore observations altogether that contain missing values, which reduces the size of the training data set. Less training data can substantially weaken the predictive power of these models. To overcome this obstacle of missing data, you can impute missing values before you fit the models.

Tip

It is a particularly good idea to impute missing values before fitting a model that ignores observations with missing values if you plan to compare those models with a decision tree. Model comparison is most appropriate between models that are fit with the same set of observations.

To use the Impute node to impute missing values, complete the following steps:

Select the Modify tab on the Toolbar.
Select the Impute node icon. Drag the node into the Diagram Workspace.
Connect the Replacement node to the Impute node.
Select the Impute node. In the Properties Panel, scroll down to view the Train properties:
- For class variables, click on the value of Default Input Method and select Tree Surrogate from the drop-down menu that appears.
- For interval variables, click on the value of Default Input Method and select Median from the drop-down menu that appears.
The default input method specifies which is the default statistic to use to impute missing values. In this example, the values of missing interval variables are replaced by the median of the nonmissing values. This statistic is less sensitive to extreme values than the mean or midrange and is therefore useful for imputation of missing values from skewed distributions. The values of missing class variables, in this example, are imputed using predicted values from a decision tree. For each class variable, SAS Enterprise Miner builds a decision tree (in this case, potentially using surrogate splitting rules) with that variable as the target and the other input variables as predictors.
In the Diagram Workspace, right-click the Impute node, and select Run from the resulting menu. Click Yes in the confirmation window that opens.
In the window that appears when processing completes, click OK.

Note: In the data that is exported from the Impute node, a new variable is created for each variable for which missing values are imputed. The original variable is not overwritten. Instead, the new variable has the same name as the original variable but is prefaced with IMP_. The original version of each variable also exists in the exported data and has the role Rejected. In this example, SES and URBANICITY have values replaced and then imputed. Therefore, in addition to the original version, each of these variables has a version in the exported data that is prefaced by IMP_REP_.