A new multi-objective evolutionary approach for creating ensemble of classifiers

Abstract

In recent years, an increasing amount of research has been focused on feature selection 
techniques. These techniques rely on an idea that by selecting the most discriminant 
features, it may reduce the number of features and increase the recognition. Instead of 
using a feature selection technique which has been widely used in multi objective 
evolutionary approaches for ensemble generating, this paper presents a new multi 
objective evolutionary algorithm based on the NSGA II which automatically preserves 
diversity and also covers problems with lower dimensional feature spaces in which 
using feature selection technique may lead to ambiguous subspaces. After creating 
classifiers based on the amount of error created for each class, another multi-objective 
genetic algorithm was used to combine them and to produce a set of powerful ensembles. 
Comprehensive experiments demonstrate the effectiveness of the proposed strategy.