Multiple Classifier System with Metaheuristic Algorithms

Ability of Multiple Classifier Systems (MCSs) to deliver correct prediction, even though all of the classifiers from its ensemble are wrong, confirms that MCS approach possess large potential in the field of pattern recognition. This work focuses on the problem of fuser design. While most MCSs use popular Voting combiner, authors of this work investigate prediction accuracy of MCS with either Evolutionary or Particle swarm algorithms applied as the fusers. Obtained results suggest, that aforementioned fusion algorithms increase accuracy performance on the evaluated datasets in comparison to the popular Voting combiner.