Rules Extracted Automatically By Induction
S. Rabaséda, R. Rakotomalala, D. A. Zighed
E.R.I.C_Lyon Université Lumière Lyon 2
5, avenue P. Mendès-France 69676 Bron Cedex
Tél : 33-4-78-77-23-20 Fax : 33-4-78-77-23-75
Among the techniques used for the automatic extraction of knowledge, the arborescent and non-arborescent processes have an important place. The results of those processes are presented under the shape of induction graphs which can then be re-expressed by production rules similar to those used in Artificial Intelligence in an expert system. So far, algorithms of automatic extraction of rules have been proposed, and exclusively based on arborescent decision processes. In this paper is presented an algorithm of automatic extraction of rules from induction graphs adapted to arborescent and non-arborescent processes. However, rewriting a graph like a simple collection of rules generates a certain number of problems. The direct use of these rules, in the knowledge base of an expert system is not possible for several reasons. On the one hand, the rules of a knowledge base are exclusively conjunctive and non conjunctive-disjunctive as those presented in induction graphs. On the other hand, the rules can convey redundant and incoherent information. In the algorithm of automatic extraction of rules, we also propose a simplification and an optimization of rules.
Key words : Numeric Learning, Induction Graph, Automatic Extraction of Rules, Simplification and Optimization of Rules.