(a) (b)

a) A data set for classification using the orthogonal partitioning rules. (b) The

rthogonal decision tree.

ver, if the oblique partitioning rule is employed, a decision tree

much simpler and more efficient. For instance, for the same data

own in Figure 3.48(a), one oblique partitioning rule can be used

te a perfect partition leading to two subspaces, each of which is

one class. Figure 3.48(b) shows such a decision tree, where only

tioning rule is employed. It is no doubt that a decision tree

d using the oblique division approach can generate a more

ious and efficient tree structure, and perhaps with a better

nce.

(a) (b)

(a) A data set for classification using an oblique partitioning rule. (b) The

decision tree.

basic principle of the random forest algorithm is to avoid

g a classifier through creating many smaller trees [Ho, 1998,

al., 2001]. Unlike a decision tree model, a random forest model

oblique hyperplanes which can have better performance with

es as the example shown in Figure 3.48. The random forest