Title:
"Towards adaptive and generic solutions for interesting pattern mining in data."
I did my PhD (2003-2006) in the University of Clermont-Ferrand (France). From december 2005 to december 2006, I was invited at the LIRIS laboratory, INSA-Lyon (Lyon, France).
My advisors were Pr. Jean-Marc Petit and Dr. Fabien De Marchi (LIRIS, Lyon, France).
I defended my PhD in december 2006 in front of:
- Pr. Alain Quilliot, ISIMA (President)
- Pr. Christine Collet, ENSIMAG (Reviewer)
- Pr. Pascal Poncelet, Ecole des Mines d'Alès (Reviewer)
- Pr. Michel Schneider, University of Clermont-Ferrand (Examiner)
- Pr. Farouk Toumani, University of Clermont-Ferrand (Examiner)
- Dr. Fabien De Marchi, University of Lyon (Supervisor)
- Pr. Jean-Marc Petit, INSA Lyon (Supervisor)
Abstract
The discovery of frequent itemsets is a famous problem in data mining. In order to better understand the influence of data on algorithms behavior, we present an experimental study of datasets commonly used by the community. This study led to a new classification of datasets according to borders distribution: stable and which contributes to explain algorithms performances. In spite of the great amount of work and a theoretical framework for interesting patterns discovery, using these algorithms to solve "equivalent" problems is difficult and sometimes impossible. Based on these limits, we proposed in this framework a generic algorithm for borders discovery, called ABS (Adaptive Borders Search), which dynamically adapts its strategy according to data. Moreover, a generic library of C++ components was proposed to facilitate the development of software solutions for this class of problems.