Section: New Software and Platforms

MO-Mine

Participants : Clarisse Dhaenens, Benjamin Fisset, Laetitia Jourdan.

The MO-Mine platform (mo-mine.gforge.inria.fr ) aims at providing optimization algorithms, and in particular multi-objective approaches, to deal with data-mining classical tasks (Classification, association rules...). The platform is still in development in the context of an Inria ADT.

$MO-Min{e}_{clust}$ is the first package of the platform and is dedicated to clustering (unsupervised classification). Indeed, it is well-known that clustering may be seen as a bi-objective optimization problem as the goal is both to minimize distances between data belonging to a same cluster, while maximizing distances between data belonging to different clusters. Several models (objective functions used,...) and engines (optimization algorithms) have been implemented. The framework searches, for a given dataset, the best association of model/engine/parameter without specifiing the number of clusters. $MO-Min{e}_{clust}$ shows very interesting behavior and shows that the choice of the model and the engine has a great importance in the performance of the method and depends on the dataset to analyze.

Inria software self-assessment for $MO-Min{e}_{clust}$: [A-1, SO-4, SM-3-up4, EM-2, SDL-1,DA-4,CD-4]