Exception Dags as Knowledge Structures

: The problem of transforming the knowledge bases of performance systems using induced rules or decision trees into comprehensible knowledgestructures is addressed. A knowledgestructure is developedthat generalizes and subsumesproductionrules, decision trees, and rules with exceptions. It gives rise to a natural complexitymeasurethat allows themto be understood, analyzedand comparedon a uniformbasis. This structure is a rooted directed acyclic graph with the semantics that nodesare concepts, someof whichhave attached conclusions, and the arcs are 'isa' inheritance links Withdisjunctive multiple inheritance. Adetailed exampleis given of the generation of a range of such structures of equivalent performancefor a simple problem,and the complexitymeasureof a particular structure is shownto relate to its perceived complexity. Thesimplest structures are generated by an algorithm that factors commonconcepts fromthe premises of rules. Amorecomplexexampleof a chess dataset is used to ...

• Type: Conference
• Year: 1994
• Where: KDD
• Authors: Brian R. Gaines
• Keywords: Data Mining, Decision Trees, Directed Acyclic Graph, KDD 1994, Withdisjunctive Multiple Inheritance