Pruning adaptive boosting ensembles by means of a genetic algorithm

Abstract

This work analyzes the problem of whether, given a classification ensemble built by Adaboost, it is possible to find a subensemble with lower generalization error. In order to solve this task a genetic algorithm is proposed and compared with other heuristics like Kappa pruning and Reduce-error pruning with backfitting. Experiments carried out over a wide variety of classification problems show that the genetic algorithm behaves better than, or at least, as well as the best of those heuristics and that subensembles with similar and sometimes better prediction accuracy can be obtained. © Springer-Verlag Berlin Heidelberg 2006.