How do Decision Trees behave with missing not at random (MNAR) data?

Decision Trees can misinterpret MNAR missingness as signal, leading to biased splits.

When data is Missing Not At Random (MNAR), the probability of missingness depends on the unobserved value itself. Decision Trees may treat missingness as informative if encoded implicitly, causing splits that exploit missing patterns rather than true feature relationships. This leads to biased models. Some implementations handle missing values via surrogate splits or dedicated missing branches, but MNAR still remains a fundamental challenge requiring domain understanding or explicit modeling.