MEMETIC PROGRAMMING WITH THE ATOMIC REPRESENTATION FOR EXTRACTING LOGICAL CLASSIFICATION RULES

Abstract

Classification is one of the most popular techniques of data mining. This paper presents an evolutionary approach for designing classifiers for two-class classification problems using an enhanced version of the genetic programming (GP) algorithm, called the Memetic Programming (MP) algorithm. MP can discover relationships between observed data and express them logically. MP aims to obtain a classifier with the largest area under the ROC curve, which has been proved a better performance than traditionally metrics. The proposed approach is being demonstrated by experimenting on some UCI Machine Learning data sets. Results obtained in these experiments reflect the efficiency of the proposed algorithm.