Mining Numeric Association Rules with Genetic Algorithms

Abstract

In this last decade, association rules are being, inside Data Mining techniques, one of the most used tools to find relationships among attributes of a database. Numerous scopes have found in these techniques an important source of qualitative information that can be analyzed by experts in order to improve some aspects in their environment. Nowadays, there are different efficient algorithms to find these rules, but most of them are demanding of databases containing only discrete attributes. In this paper we present a tool, GENAR (GENetic Association Rules), that discover association rules in databases containing quantitative attributes. We use an evolutionary algorithm in order to find the different intervals. We also make use of the evolutionary methodology of iterative rule learning to not evolve always to the same rule. By means of this we get to discover the different association rules. In our approach we present a tool that obtain association rules with an undetermined number of numeric attributes in the antecedent of the rule.