On the Discovery of Exception Rules: A Survey

Béatrice Duval (1) , Ansaf Salleb (2, 3) , Christel Vrain (4)

1 LERIA - Laboratoire d'Etudes et de Recherche en Informatique d'Angers
2 DREAM - Diagnosing, Recommending Actions and Modelling
3 CCLS - Center for Computational Learning Systems
4 LIFO - Laboratoire d'Informatique Fondamentale d'Orléans

In this paper, we present a survey of the different approaches developed for mining exception rules. Exception rules are interesting in the context of quality measures since such rules are intrinsically satisfied by few individuals in the database and many criteria relying on the number of occurrences, such as for instance the support measure, are no longer relevant. Therefore traditional measures must be coupled with other criteria. In that context, some works have proposed to use the expert's knowledge: she/he can provide the system either with constraints on the syntactic form of the rules, thus reducing the search space, or with commonsense rules that have to be refined by the data mining process. Works that rely on either of these approaches, with their particular quality evaluation are presented in this survey. Moreover, this presentation also gives ideas on how numeric criteria can be intertwined with user-centered approaches.

Domaines: Apprentissage [cs.LG]

Béatrice Duval, Ansaf Salleb, Christel Vrain. On the Discovery of Exception Rules: A Survey. F. Guillet, H.J. Hamilton. Quality Measures in Data Mining Book, Springer in the Series Studies in Computational Intelligence, pp.77-98, 2007. ⟨hal-00084863⟩