Skip to Main content Skip to Navigation
Theses

Techniques d'optimisation pour la fouille de données

Abstract : Numerical technologies have generated for a few years, huge volumes of data, which can conceal useful information. This situation gave rise to knowledge discovery activities in data bases. This indicates the non obvious process of extracing implicit informations, previously unknown and potentially useful hidden into the data. A standard knowledge discovery process includes five steps. The main one is data mining. We are interested in a kind of information expressed as a dependency rule and in the interestingness of a rule. A dependency rule is a conditional implication between sets of attributes over the data set. The purpose of standard data mining algorithmes is generally to find the best model. In fact, behind these processes, there is an optimization problem which is not explicitly expressed. We consider interestingness of dependency rules as being an optimization problem in which rule interestingness is quantified by the mean of measures. Thus, it is necessary to study the search space induced by measures as well as seach algorithms associated with the analysis of these spaces. It arises that these measures have a different behavior according to the data set involved ; so, an analytical approach is not possible. It arises that some of these measures, when they are considered simultaneously, present antagonisms ; thus, obtaining "the" best rule is not possible ; it is necessary to consider a set of good tradeoffs. We bring solutions by the means of genetic approch.
Document type :
Theses
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-00216131
Contributor : Estelle Nivault <>
Submitted on : Thursday, January 24, 2008 - 3:29:51 PM
Last modification on : Monday, October 12, 2020 - 10:30:21 AM
Long-term archiving on: : Thursday, April 15, 2010 - 2:44:26 AM

Identifiers

  • HAL Id : tel-00216131, version 1

Collections

Citation

Dominique Francisci. Techniques d'optimisation pour la fouille de données. Interface homme-machine [cs.HC]. Université Nice Sophia Antipolis, 2004. Français. ⟨tel-00216131⟩

Share

Metrics

Record views

410

Files downloads

609