Enhancing difficult classes in one-vs-one classifier fusion strategy using restricted equivalence functions

One-vs-One is a commonly used decomposition strategy to overcome multi-class problems, even when the base classifier supports directly addressing the multi-class problem. This paper analyzes the fact that, in this strategy, less attention is given to the difficult classes, favoring the easier ones. Different evaluation criteria are used, and a novel fusion strategy, which generalizes the weighted voting, is presented to enhance the difficult classes classification. The new methodology is able to increase the recognition of the difficult classes, thus obtaining a more balanced performance over all classes, which is a desirable behavior.