文档介绍：Discovering Interesting Prediction Rules with a ic Algorithm
Edgar Noda 1, Alex A. Freitas 2 and Heitor S. Lopes 3
1,3 CEFET-PR 2 PUC-PR
CPGEI. ET.
Av. 7 de Setembro, 3165. R. Imaculada Conceicao, 1155
Curitiba - PR. Curitiba – PR.
80230-901. Brazil -901. Brazil
******@, ******@
******@ .br/~alex
Abstract- In essence, the goal of data mining is to This paper proposes a ic Algorithm (GA) designed
discover knowledge which is highly accurate, to discover interesting prediction rules. The proposed
comprehensible and “interesting”(surprising, novel). bines some characteristics of the GA-Nuggets
Although the literature emphasizes predictive accuracy algorithm [Freitas, 99] with some ideas on how to evaluate
prehensibility, the discovery of interesting rule interestingness in an objective (data-driven, domain-
knowledge remains a formidable challenge for data independent) manner [Freitas, 98].
mining algorithms. In this paper we present a ic Our GA was designed for dependence modelling, a data
algorithm designed from the scratch to discover mining task which can be seen as a generalization
interesting rules. Our GA addresses the dependence of the classification task. In this latter the aim is to
modelling task, where different rules can predict predict the value of a special attribute, called the goal
differe