Hybrid and Traditional Metaheuristic Optimization for Desgining Ensemble of Classifiers
Ensemble of Classifiers; Metaheuristic Optimization
This work discusses the use of traditional and hybrid optimization techniques in the
context mono and multi-objective for optimization of the problem of pattern classification
in ensembles. The problem of pattern classification is treated as a problem of optimization
looking to find the subset of attributes and classifiers of problem that minimizes the classification
error of the ensemble. Experiments are performed in different scenarios mono and
multi-objective optimizing classification error and the measures of good and bad diversity.
The objective is to verify if you add the diversity measures as optimization objectives
results in more accurate ensemble. Thus, the contribution of this study is to determine
whether the measures of good and bad diversity can be used on optimization techniques
mono and multi-objective as optimization objectives for construction of committees of
more accurate classifiers that those built by the same process but using only the accuracy
of classification as objective of optimization.