The example above demonstrates several differences resulting
from the use of hierarchical class labels. First, in selecting a splitting
attribute to grow the tree, we must consider not only accuracy,
but also class specificity. Second, in deciding when to stop
growing a node and assign a label to the node, the distribution
of data over the class hierarchical tree must be considered in order
to determine a proper label. This paper proposes a novel
classification algorithm to accommodate these differences by
taking into account the distribution of class labels in the hierarchical
tree so that the precision and accuracy of a DT can be
optimized.