An Ensemble Co-Evolutionary based Algorithm for Classification Problems

Abstract

In this paper, the authors propose a dual-population co-evolutionary approach using ensemble learning approach (E-SOCA)  to  simultaneously  solve  both  feature  subset selection  and  optimal  classifier  design.  Different  from previous  studies  where  each  population  retains  only  one best individual (Elite) after co-evolution, in this study, an elite  community  will  be  stored  and  calculated  together through  an  ensemble  learning  algorithm  to  produce  the final    classification    result.    Experimental    results    on standard  UCI  problems  with  a  variety  of  input  features ranging from small to large sizes shows that the proposed algorithm  results  in  more  accuracy  and  stability  than traditional algorithms.