Análise de novas abordagens para mineração de regras de classificação utilizando algoritmos genéticos

Abstract

The classification task is among the most used in Data Mining and is widely researched nowadays. Several works have already been developed using Genetic Algorithms for classification tasks through the evolution of IF-THEN classification rules and good results have been obtained. These methods often use the same chromosome structure, integer, and real values, and this structure may impose some limitations on their operation. Also, these methods use the same approach for choosing the attributes that will compose the rules. In this work, two new methods were proposed, called BIN-NLCEE and IG-CEE. BIN-NLCEE uses a new chromosomal structure by binary values. The IG-CEE method uses an attribute evaluation measure, called Information Gain, to select the attributes that can compose the rules. Four medical domain datasets were used for BIN-NLCEE validation and 3 synthetic datasets for IG-CEE validation. Both were compared with their source methods and 4 other traditional classifiers (J48, IBK, Naive Bayes and SVM). The results showed that the proposed methods were able to generate better fitness values and better convergence rates.