Up to until now, DT construction algorithms have usually assumed
that the class labels were categorical or Boolean variables,
meaning that the algorithms operate under the assumption that
the class labels are flat. In real-world applications, however, there
are more complex classification scenarios, where the class labels to
be predicted are hierarchically related. For example, in a digital library
application, a document can be assigned to topics organized
into a topic hierarchy (Wu, Zhang, & Honavar, 2005); in web site
management, a web page can be placed into categories organized
as a hierarchical catalog. In both of these cases, the class labels
are naturally organized as a hierarchical structure of class labels
which defines an abstraction over class labels. Unfortunately, existing
research has paid little attention to the classification of data
with hierarchical class labels. To the best of our knowledge, no
method has been developed to construct DTs directly from data
with hierarchical class labels. To remedy this research gap, this
work proposes such an algorithm.