Pairwise classifier combination in the transferable belief model
Benjamin Quost, T. Denaeux, Marie-Hélène Masson · 2005
Classifier combination constitutes an interesting approach when solving multi-class classification problems. We propose to carry out this combination in the belief functions framework. Our approach, similar to a method proposed by Hastie and Tibshirani in a probabilistic framework, is first presented. The performances obtained on various datasets are then analyzed, showing a gain of classification accuracy using the belief functions approach.