Hybridizing evolutionary algorithms for creating classifier ensembles

• Published in: 2014 Sixth World Congress on Nature and Biologically Inspired Computing (NaBIC 2014) pp. 84 - 90
• Main Authors: Dufourq, Emmanuel, Pillay, Nelishia
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01-07-2014
• Subjects: Classification algorithms; Clustering algorithms; data classification; data mining; ensemble classifiers; genetic algorithms; genetic programming; Genetics; Meteorology; Sociology; Sonar; Statistics
• DOI: 10.1109/NaBIC.2014.6921858

Genetic programming (GP) has been applied to solve data classification problems numerous times in previous studies and the findings in the literature confirm that GP is able to perform well. In more recent studies, researchers have shown that using a team of classifiers can outperform a single classifier. These teams are referred to as ensembles. Previously, several different attempts at creating ensembles have been investigated; some more complex than others. In this study, four approaches have been proposed, in which the ensemble methods hybridize a genetic algorithm with a GP algorithm in different ways. The first three approaches made use of a generational GP model, while the fourth used a steady state GP model. The four approaches were tested on eight public data sets and the findings confirm that the proposed ensembles outperform the standard GP method, and additionally outperform other GP methods found in literature.