What should you do if your input variables have missing values before running a decision tree?

Choosing not to impute any missing variables because trees can handle them is the appropriate action in this context. Decision trees, such as those used in SAS Enterprise Miner, have a unique advantage when managing missing data; they can incorporate the missing values in the node-splitting process. This means that decision trees can effectively work with records that have missing values without requiring imputation.

When a decision tree encounters a missing value for an input variable, it can continue its decision-making process by considering the instances that either do not have a value for that variable or by splitting the data in a way that accommodates the missing information, allowing the tree to still create meaningful splits based on available variables. This inherent flexibility is one of the reasons why decision trees remain a popular choice among predictive modeling techniques.

In contrast, imputing missing values for all variables can lead to biases or inaccuracies in the model, especially if the imputation method does not appropriately reflect the underlying data distribution. Removing all variables with missing values could lead to loss of potentially valuable information and reduce the dataset's size, which may hurt model performance. Therefore, the strategy of allowing the decision tree to handle missing values is both efficient and effective.