Tree-structured ranking rules and approximation of the optimal ROC curve

Abstract : We consider the extension of standard decision tree methods to the bipartite ranking problem. In ranking, the goal pursued is global: define an order on the whole input space in order to have positive instances on top with maximum probability. The most natural way of ordering all instances consists in projecting the input data x onto the real line using a real-valued scoring function s and the accuracy of the ordering induced by a candidate s is classically measured in terms of the AUC. In the paper, we discuss the design of tree-structured scoring functions obtained by maximizing the AUC criterion. In particular, the connection with recursive piecewise linear approximation of the optimal ROC curve both in the L_1 and in the L_{\infty}-sense is discussed.
Type de document :
Pré-publication, Document de travail
Liste complète des métadonnées
Contributeur : Nicolas Vayatis <>
Soumis le : vendredi 4 avril 2008 - 00:50:29
Dernière modification le : jeudi 11 janvier 2018 - 06:23:38
Document(s) archivé(s) le : mardi 21 septembre 2010 - 15:59:22


Fichiers produits par l'(les) auteur(s)


  • HAL Id : hal-00268068, version 2


Stéphan Clémençon, Nicolas Vayatis. Tree-structured ranking rules and approximation of the optimal ROC curve. 2008. 〈hal-00268068v2〉



Consultations de la notice


Téléchargements de fichiers