DOCUMENTATION
Sign In

Get access to My SAS, trials, communities and more.

Machine Learning Programming Guide
 
CASL
Lua
Python
R

Explain Model Action Set

Handling Missing Data

You can control the handling of missing data in the linearExplainer action by specifying the includeMissing parameter and the impute subparameter of the dataGeneration parameter. By default, the includeMissing parameter is set to False and the impute subparameter is set to NONE.

The includeMissing parameter specifies whether to consider a missing value of a nominal variable to be a valid level. This parameter does not affect the handling of missing data in interval variables. It takes the value False or True:

  • False indicates that a missing value of a nominal variable is not considered to be a valid level. The action ignores missing values when calculating the distances. The action does not generate new data that have missing values of the interval or nominal variables.

  • True indicates that a missing value of a nominal variable is considered to be a valid level. If the query data have a missing nominal variable, this variable is considered in the distance calculation, where the hamming distance is 0 only if the input reference data also have a missing value of the corresponding variable. The action can also generate new data that have missing values of nominal variables, if the input reference data table has missing values of the corresponding nominal variables. The proportion of the missing values is determined by the reference data table. Interval variables are still generated without missing values.

The impute subparameter specifies how the linearExplainer action imputes a missing value. When the includeMissing parameter is set to False, the impute subparameter controls the imputation behavior in both the interval and nominal variables. When the includeMissing parameter is set to True, the impute subparameter controls the imputation behavior in only the interval variables. The impute subparameter takes the value MEAN, NONE, or WEIGHTEDMEAN:

  • MEAN indicates that the linearExplainer action imputes missing values in the input reference data and the query data by using the variable mean or the variable mode of the input reference data when applicable. If the includeMissing parameter is set to False, the action replaces missing values of an interval variable with the mean value of the corresponding variable, and it replaces missing values of a nominal variable with the most frequent level of the corresponding variable. If the includeMissing parameter is set to True, the action replaces only missing values of an interval variable with the mean value of the corresponding variable. Note that although the action imputes the missing values in the query data before sending the data to the glm action, the distance calculation is done between the imputed input data and the original query data without imputation. This means that if the query data have a missing value of a variable and the missing values of this variable are not considered to be a valid level, the action ignores the variable when calculating the distance.

  • NONE indicates that the linearExplainer action does not impute missing values in the input data or the query data.

  • WEIGHTEDMEAN indicates that the linearExplainer action imputes missing values in the reference data and the query data by using the weighted variable mean or the variable mode of the input reference data, whichever is applicable. The action use the same imputation process as for the MEAN value, except that it uses the weighted variable mean for interval variables.

Last updated: August 04, 2026