LIP6 2002/022
- Thesis
Bases de données multidimensionnelles floues et leur utilisation pour la fouille de données - A. Laurent
- 231 pages - 09/20/2002- document en - http://www.lip6.fr/lip6/reports/2002/lip6.2002.022.pdf - 4,739 Ko
- Contact : Anne.Laurent (at) nulllip6.fr
- Ancien Thème : APA
- Keywords : Multidimensional Databases, OLAP, Data Mining, Fuzzy Logic, Fuzzy Summaries
- Publisher : Ghislaine.Mary (at) nulllip6.fr
Since the early 1990s, data warehousing has provided an efficient framework to deal with huge amounts of data in an analytical view. It has appeared that transactional systems (OLTP - On-Line Transactional Processing) are not suitable for fast multidimensional analysis. For this reason, multidimensional databases have emerged to support OLAP tools (On-Line Analytical Processing). However, real world data are often imperfect, being either uncertain, or imprecise. This explains why it is important for database management systems to be able to cope with flexible queries. In this framework, fuzzy set theory is appropriate for improving information systems in particular for qualitative interpretation of quantitative data (numerical data). The information provided to users (for instance by means of rules) is more understandable, and generalizable. Our work aims at defining an approach to perform OLAP-based mining using fuzzy multidimensional databases and fuzzy data mining algorithms. We propose an extension of multidimensional databases to handle imperfect information and flexible multidimensional queries. We also integrate fuzzy multidimensional databases with machine learning methods. More specifically, using the proposed model, we introduce a system that allows to build fuzzy decision trees and to generate linguistic summaries using the proposed data model. A general architecture is provided, based on fuzzy multidimensional databases as a support for knowledge discovery. We also study from a semantic point of view the problem of abnomalies related to the presence of empty cells.