It exploits the theoretical basis and attractive
features of formal concept analysis and concept lattices to propose two techniques:
(1) one which exploits lattice based mining algorithms for efficiently generating
frequent closed itemsets from multidimensional data to further extract association
rules and clusters, and (2) the second technique which defines new operators similar
in spirit to OLAP techniques to allow ???data mining on demand??? by relying on
operations like projection, selection, and assembly on concept lattices.
2 4 Missaoui, Jatteau, Boujenoui, & Naouali
Copyright ?© 2007, Idea Group Inc. Copying or distributing in print or electronic forms without written permission of
Idea Group Inc. is prohibited.
Our current work covers also statistical modeling in data warehouses in order to
discover useful patterns in data (e.g., outliers), discard irrelevant dimensions and
dimension members, and hide irrelevant cube cells. In particular, we are exploring
the potential of log-linear modeling for data summarization and prediction of multidimensional
data. Among the observed barriers, we note the difficulty to efficiently
choose a parsimonious model (i.e., a reduced model that fits the data) from a possibly
very large set of candidate models, and the applicability of log-linear models
to high-dimension cubes.
Acknowledgments
We would like to thank the Natural Sciences and Engineering Research Council
of Canada (NSERC) and Le fonds qu?©b?©cois de la recherche sur la nature et les
technologies (FQRNT) for their financial support.
Pages:
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507