A multi-objective competitive co-evolutionary approach for classification problems

This paper proposes a multi-objective competitive co-evolutionary algorithm (MOCPCEA) based on the PreyPredator model to solve classification problems. In the MOCPCEA, a data population acts as preys. To be specific, each prey represents a selected subset of the training dataset. Another population is ANN classifiers which play as Predators. The task of the Predators is to try to classify the data sets as correctly as possible, whereas the Preys try to find the data sets that are difficult to be classified. Through this interaction process, MOCPCEA generates a set of classifiers that are able to classify difficult data sets. The final classification result is given by the ensemble voting mechanism among these sets of classifiers. The performance of the proposed algorithm is performed on seven benchmark problems. Through comparison with other algorithms, the proposed algorithm indicates that it could create an ensemble of ANN networks that give high and stable classification results.