Experimental evaluation of two new GEP-based ensemble classifiers

► Gene Expression Programming is used to define two new ensemble classifiers. ► Two quality measures are proposed and used in gene selection. ► Validation experiments were performed and results compared with other methods. ► Both classifiers give good classification accuracy. ► Both classifiers are...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Expert systems with applications Ročník 38; číslo 9; s. 10932 - 10939
Hlavní autoři: Je, Joanna, Drzejowicz, Drzejowicz, Je, Piotr
Médium: Journal Article
Jazyk:angličtina
Vydáno: Elsevier Ltd 01.09.2011
Témata:
Algorithms
Classification
Classifiers
Expert systems
Gene expression
Gene Expression Programming
Genes
Programming
Voting
Gene Expression Programming
Classification
ISSN:0957-4174, 1873-6793
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:► Gene Expression Programming is used to define two new ensemble classifiers. ► Two quality measures are proposed and used in gene selection. ► Validation experiments were performed and results compared with other methods. ► Both classifiers give good classification accuracy. ► Both classifiers are competitive in terms of area under ROC curve. The paper proposes applying Gene Expression Programming (GEP) to induce ensemble classifiers. Two new algorithms inducing such classifiers are proposed. The proposed ensemble classifiers use two different measures to select genes produced by the Gene Expression Programming procedure. Selection of genes from the set of the non-dominated ones in the process of meta-learning is supported by a genetic algorithm. Integration of genes (i.e. learners) is based on the majority voting. The proposed algorithms were validated experimentally using several datasets and the results were compared with those of other well established classification methods.
Bibliografie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:0957-4174
1873-6793
DOI:10.1016/j.eswa.2011.02.135