Tree-Based Machine Learning Methods in SAS Viya

Chapter 4: Decision Trees: Strengths, Weaknesses, and Uses

Missing Values in Decision Trees

You might have to contend with missing values among the inputs as shown by question marks in Figure 4.1. Decision trees accommodate missing values very well compared to other modeling methods. Decision trees that split on one input at a time are more tolerant to missing data than models such as regression that combine several inputs. In regression models, an observation missing any input value is discarded (complete case analysis).

For the simplest of tree algorithms, only observations that need to be excluded are those missing the input currently being considered to split on. They can be included when considering splitting on a different input (for example, ...