Decision trees using class hierarchy
Tomoya Takamitsu, Takao Miura, Isamu Shioya · Hybrid Intelligent Systems · 2003
In this work, we propose Disjunctive Decision Trees to obtain simple, reliable and interesting decision trees. To do that we introduce domain knowledge in a form of class hierarchy and we relax class membership. Eventually we lose small amount of entropy. This is why we define path entropy to evaluate interests of decision trees. We discuss some experimental results and show how useful these trees are.