Genetic rule selection with a multi-classifier coding scheme for ensemble classifier design

Abstract

In this paper, we examine the effectiveness of genetic rule selection with a multi-classifier coding scheme for ensemble classifier design. Genetic rule selection is a two-stage method. The first stage is rule extraction from numerical data using a data mining technique. Extracted rules are used as candidate rules. The second stage is evolutionary multiobjective rule selection from the candidate rules. We use a multi-classifier coding scheme where an ensemble classifier is represented by an integer string. Three criteria are used as objective functions in evolutionary multiobjective rule selection to optimize ensemble classifiers in terms of accuracy and diversity. We examine the performance of designed ensemble classifiers through computational experiments on six benchmark datasets in the UCI machine learning repository.