I was reading the limitations of decision trees. One of them was that, for classification problems, decision trees produce only orthogonal decision boundaries. Could anyone please explain what an orthogonal decision boundary is?
Often, every node of a decision tree creates a split along one variable - the decision boundary is "axis-aligned". The figure below from this survey paper shows this pictorially. (a) is axis-aligned: the decision boundary uses variable $x_1$ only. (b) is not axis-aligned: it uses both input variables, but is linear. (c) is non-linear and not axis-aligned.
In what you read, "orthogonal decision boundary" probably means the same thing as "axis-aligned decision boundary".