Advanced Decision Trees - Data Mining with Decision Trees: Theory and Applications - page 173

the instance space using only axis-parallel separating surfaces (typically,
hyperplanes). Several cases presented in the literature use multivariate
splitting criteria.
In multivariate splitting criteria, several attributes may participate in
a single node split test. Obviously, finding the best multivariate criteria
is more complicated than finding the best univariate split. Furthermore,
although this type of criteria may dramatically improve the trees perfor-
mance, these criteria are much less popular than the univariate criteria.
Figure 11.6 presents a typical algorithmic framework for top-down
inducing of oblique decision trees. Note that this algorithm is very similar
Fig. 11.6 Top-down algorithmic framework for oblique decision trees induction.