Skip to Main content Skip to Navigation
Theses

Algorithmes métaheuristiques hybrides pour la sélection de gènes et la classification de données de biopuces

Abstract : DNAmicroarray technologies permit tomeasure simultaneously gene expressions for thousands of genes in a sample and enable to consider molecular cancer diagnosis based on gene expression. Data that are currently available in this field concern a very large number of variables (thousands of gene expressions) relative to a small number of observations (typically under one hundred samples). This thesis deals with the problem of gene selection, which aims to propose a subset of relevant genes in order to build efficient classifiers to recognize different types of tumor. The problem of gene selection is a very hard problem, for which metaheuristics algorithms based on neighbourhood (local search methods) and population (genetic algorithms and memetic algorithms) seem appropriate. In this thesis, we propose several embedded gene selection methods, that combine metaheuristics algorithms with a support vector machine. In these algorithms, the quality of a selected gene subset is evaluated by a linear SVM classifier trained on this subset. Moreover, these algorithms use the relevance measure, given by the linear SVM about each gene, to inform the search process or to build very specialized genetic operators. Experimentations performed on available data sets show very competitive results when compared to the state-ofthe-art works.
Document type :
Theses
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-00447684
Contributor : Anne-Marie Plé <>
Submitted on : Friday, January 15, 2010 - 3:39:17 PM
Last modification on : Tuesday, May 7, 2019 - 6:30:13 PM
Long-term archiving on: : Thursday, June 17, 2010 - 10:50:42 PM

Identifiers

  • HAL Id : tel-00447684, version 1

Citation

José Crispin Hernandez Hernandez. Algorithmes métaheuristiques hybrides pour la sélection de gènes et la classification de données de biopuces. Informatique [cs]. Université d'Angers, 2008. Français. ⟨tel-00447684⟩

Share

Metrics

Record views

523

Files downloads

3861