‘Entropy’ on Covers and Its Application on Decision TreeConstruction
Zhimin Wang · International Journal of Machine Learning and Computing · 2011
Decision tree is a popular classification tool. To automatically construct a good decision tree, people have introduced entropy as a heuristic for attribute selection to deal with the intractable nature of finding an optimal solution with regard to the size of a tree. To solve a special kind of decision tree construction used in biological taxonomy, we need consider polymorphic attributes, against which a single instance may hold different values. To properly evaluate polymorphic attributes during tree construction, we propose the conditional form of a novel 'entropy' measure called 'disconnectivity' as the heuristic. In parallel to the theory of generalized entropy, 'disconnectivity' is also generalized to a family of measures.