What is surrogate splitting in Decision Trees?

Surrogate splitting is a technique where alternative features are used to replicate the behavior of a primary split when data is missing.

Surrogate splits are used in algorithms like CART to handle missing values during training and prediction. When the primary splitting feature is missing, the model uses another feature that best mimics the original split's partitioning. These surrogate features are ranked by how well they agree with the original split. This allows decision trees to maintain structure without discarding incomplete data, improving robustness in real-world datasets with missing values.