Toward high performance solution retrieval in multiobjective clustering
View/Open
Author
García Piquer, Álvaro
Sancho Asensio, Andreu
Fornells Herrera, Albert
Golobardes, Elisabet
Corral Torruella, Guiomar
Teixidó Navarro, Francesc
Other authors
Universitat Ramon Llull. Facultat de Turisme i Direcció Hotelera Sant Ignasi
Universitat Ramon Llull. La Salle
Institut de Ciències de l'Espai
Publication date
2019-10Abstract
The massive generation of unlabeled data of current industrial applications has attracted the interest of data mining practitioners. Thus, retrieving novel and useful information from these volumes of data while decreasing the costs of manipulating such amounts of information is a major issue. Multiobjective clustering algorithms are able to recognize patterns considering several objective function which is crucial in real-world situations. However, they dearth from a retrieval system for obtaining the most suitable solution, and due to the fact that the size of Pareto set can be unpractical for human experts, autonomous retrieval methods are fostered. This paper presents an automatic retrieval system for handling Pareto-based multiobjective clustering problems based on the shape of the Pareto set and the quality of the clusters. The proposed method is integrated in CAOS, a scalable and flexible framework, to test its performance. Our approach is compared to classic retrieval methods that only consider individual strategies by using a wide set of artificial and real-world datasets. This filtering approach is evaluated under large data volumes demonstrating its competence in clustering problems. Experiments support that the proposal overcomes the accuracy and significantly reduces the computational time of the solution retrieval achieved by the individual strategies
Document Type
Article
Submitted version
Pages
33 p.
Publisher
Elsevier
Is part of
Information Sciences, 2015, Vol. 320, No. 1 (November)
This item appears in the following Collection(s)
Rights
© Elsevier. Tots els drets reservats