On Discovering Qualitative Knowledge in Rule Based Knowledge Bases: An Intelligent Approach

Production rules, also called antecedent-consequent rules, are a common way to express a knowledge base (KB) for reactive agents. Only one of the KB's rules can be triggered at any specific moment in a reactive KB. This means that an order for evaluating the rules must be established. A top-down order is typically considered, where rules at the top of the KB have a higher priority. There are more efficient ways for describing actions in terms of attribute-value pairs than production rules. In this way, decision trees constitute a simpler alternative decision structure. A decision tree fits a KB if it strictly respect the description of the actions in terms of the attributes just like the rules of the KB do. This paper describes a method to discover qualitative knowledge from a KB by generating a fitting and optimal decision tree from it. Whereas the consideration of the fitting factor for a decision tree depends exclusively on the KB from which it is induced, the optimality factor depends on the nature of the problem in hand. Thus, the resulting tree strictly respects all the properties and priorities of the KB's rules as well as the optimality criterion.