An Advanced Genetic Approach for Stacking Classifiers

Abstract

A key to obtain a high quality of stacking ensemble is to select a proper configuration for the specific dataset. Various approaches are proposed on this topic. GA-Ensemble is an effective approach using genetic algorithm to select the configuration. However, the weakness of the problem encoding and the reproduction of unnecessary individuals in GA-Ensemble impact the accuracy of the approach. In this work, the subspace partition and a tabu strategy are used to improve the accuracy of GA-Ensemble. The results over 13 UCI datasets show that the new proposed approach has a better performance.