3.1. Attribute selection measure
An attribute selection measure is a heuristic to measure which
attribute is the most discriminatory to split a current node. Many
algorithms for DT induction, such as ID3 (Quinlan, 1986) and
C4.5 (Quinlan, 1993), are suitable for problems with flat nominal
classes, and the entropy-based approach is probably the most popular
approach to selecting the most discriminatory attribute for
splitting.
In view of the weakness of the traditional entropy-based method
discussed above, we propose a new measure, called hierarchicalentropy
value, by modifying the traditional entropy measure. It can
help measure the appropriateness of a node with respect to the given
class hierarchical tree.