On multi-class learning through the minimization of the confusion matrix norm - Archive ouverte HAL Accéder directement au contenu
Rapport Année : 2013

On multi-class learning through the minimization of the confusion matrix norm

Résumé

In many multi-class classification problems, the misclassification rate as an error measure is not the relevant choice, think of the imbalanced classes problems. In order to overcome this shortcoming, several methods have been proposed where the error measure embeds richer informations than the mere misclassification rate. Yet, to the best of our knowledge, none of these methods makes use of one of the most natural tools in the multi-class setting: the confusion matrix. Recent results show that using the norm of the confusion matrix as an error measure can be quite interesting due to the additional informations contained in the matrix, especially in the case of imbalanced classes. In this paper, we show step by step how to obtain a boosting-based method which minimizes the norm of the confusion matrix. The experimental results point out that the proposed method performs better than AdaBoost.MM on imbalanced datasets, while both methods are equivalent on balanced datasets.
Fichier principal
Vignette du fichier
main_combo.pdf (190.76 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-00801313 , version 1 (15-03-2013)
hal-00801313 , version 2 (01-11-2013)

Identifiants

Citer

Sokol Koço, Cécile Capponi. On multi-class learning through the minimization of the confusion matrix norm. 2013. ⟨hal-00801313v1⟩
181 Consultations
662 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More