An Ensemble Co-Evolutionary based Algorithm for Classification Problems
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.