Extended genetic algorithm for tuning a multiple classifier system

A widely accepted idea in the pattern recognition field is that a multiple classifier system use to show superior performance than individual classifiers when dealing with complex problems. Most multiple classier systems are built up from classifiers developed completely independent of each other and combined in a last step, generating possible decisions conflicts among individual classifiers. In this paper, a non standard genetic algorithm for tuning multiple classifier systems is presented. This algorithm is based on a set of concepts that extends the genetic metaphor: coevolutionary diversity, collective fitness, suitable behavior, phylogenetic evolution and ontogenetic evolution.