Aplicação de algoritmos genéticos para a mineração de regras SE-ENTÃO em conjuntos de dados com distribuição Não-Linear

Abstract

The present work has the purpose of accomplishing the mining of high-level classification rules (IF-THEN) in data set with continuous attributes and non-linear data distribution, in a precise and comprehensible way. Thus, we used the Genetic Algorithm (GA) technique - computational research methods based on Darwin's evolutionary theory - to develop two algorithms, MIRGA and MDRGA. The MIRGA is an extension of the NLCEE algorithm of Amaral and Hruschka, and its results were similar and even inferior in comparison to the NLCEE. Consequently, the second algorithm, MDRGA, was originated and it has obtained the best classification result for the data sets, Sonar, Iris, Bupa and Ionosphere, in comparison to other methods based on GA (CEE and NLCEE) and traditional classification methods, such as: J48, Random Forest, PART, Naive Bayes, IBK and MLP.