Game-Theory Based Voting Schemas for Ensemble of Classifiers

In this paper, we present the idea of adapting the game theory concepts to the problem of estimating the weights in voting schema for ensemble methods. Such an approach provides a novel perspective, simultaneously allowing the comparison of all available classifiers. We consider a single classifier as the rational decision-maker with a unique strategy. Thus, computing the Nash equilibrium leads to deriving stable and optimal weights, maximizing the classification quality performed by the ensemble method. We compare our proposed idea with the popular voting schema and show that our concept derives similar results. At the same time, there is no need to adjust the optimal weights for the ensemble methods separately. Numerical experiments were performed on the well-known binary classification problems for ensemble methods with differing numbers of classifiers.