Mining classification rules without support: an anti-monotone property of jaccard measure

Yannick Le Bras (1, 2) , Philippe Lenca (3, 1) , Stéphane Lallich (4)

1 LUSSI - Département Logique des Usages, Sciences sociales et Sciences de l'Information
2 Lab-STICC - Laboratoire des sciences et techniques de l'information, de la communication et de la connaissance (UMR 3192)
3 Lab-STICC_TB_CID_DECIDE
4 ERIC - Equipe de Recherche en Ingénierie des Connaissances

We propose a general definition of anti-monotony, and study the anti-monotone property of the Jaccard measure for classification rules. The discovered property can be inserted in an Apriori-like algorithm and can prune the search space without any support constraint. Moreover, the algorithm is complete since, it outputs all interesting rules with respect to the measure of Jaccard. The proposed pruning strategy can then be used to efficiently find nuggets of knowledge.

Mots clés: Anti-monotony property; Jaccard measure; Classification rules

Domaines: Apprentissage [cs.LG]; Base de données [cs.DB]; Intelligence artificielle [cs.AI]; Machine Learning [stat.ML]

Yannick Le Bras, Philippe Lenca, Stéphane Lallich. Mining classification rules without support: an anti-monotone property of jaccard measure. DS 2011: 14th international conference on Discovery Science, Nov 2011, Espoo, Finland. pp.179 - 193, ⟨10.1007/978-3-642-24477-3⟩. ⟨hal-00632774⟩