Abordagens heurísticas para problemas de agrupamentos / Heuristics approaches for clustering problems


IBICT - Instituto Brasileiro de Informação em Ciência e Tecnologia




The clustering problems arose from the need to group data in order to understand an object or a phenomenon still unknown. Data clustering is based on similarity between objects of a data set, where the most similar objects are in the same group. This work proposes three new heuristics approaches to clustering problems: the \textit{Variable Neighborhood Search} (VNS) metaheuristic, the \textit{Iterated Local Search} (ILS) metaheuristic and the hybrid method called \textit{Clustering Search}(CS). VNS is characterized by performing searches in a distant neighborhoods. ILS performs a perturbation in a solution, generating new starting solutions to the local search. CS, called hybrid method because it uses a combination of metaheuristics with local search, is characterized by performing searches in promising regions to the solution improvement. In this work, these algorithms will use the clique partitioning approach to perform the clustering. The clusters obtained by these algorithms will be evaluated by two external validation indexes: Rand and Corrected Rand. Moreover, a little application in image classification will be presented. At the end, the results obtained will be compared with others algorithms in the literature.


clique partitioning metaheuristics vns ils cs ils cs clustering vns agrupamentos particionamento em cliques meta-heurísticas

Documentos Relacionados