Combining Multiple Pairwise Neural Networks Classifiers: A Comparative Study
Olivier Lézoray, Hubert Cardot · 2005
Abstract. Classifier combination constitutes an interesting approach when solving multiclass classification problems. We review standard methods used to decode the decomposition generated by a one-against-one approach. New decoding methods are proposed and are compared to standard methods. A stacking decoding is also proposed and consists in replacing the whole decoding by a trainable classifier to arbiter among the conflicting predictions of the binary classifiers. Substantial gain is obtained on all datasets used in the experiments. 1