Informática

URI Permanente desta comunidade

http://repositorio.ufes.br/handle/10/17625
Programa de Pós-Graduação em Informática

Centro: CT
Telefone: (27) 4009 2324 R*5126
URL do programa: http://www.informatica.ufes.br/pt-br/pos-graduacao/PPGI

Navegar

Navegando Informática por Autor "Alckmin, Danuza Prado de Faria"

Agora exibindo 1 - 1 de 1
  • Nenhuma Miniatura disponível
    Item
    Algoritmo genético híbrido aplicado ao problema de agrupamento de dados
    (Universidade Federal do Espírito Santo, 2009-08-31) Alckmin, Danuza Prado de Faria; Varejão, Flávio Miguel; Martins, Simone de Lima; Boeres, Maria Claudia Silva
    Clustering is a task that divides a data set in subgroups aiming that elements associated to one exactly group are more similar between themselves than elements of other groups. Organizing data in groups make it possible to identify similarities and differences between them, to extract useful information and conclusions regarding the data features. Clustering may be considered an optimization problem because it is intended to find the best combination of partitions among all possible combinations. An approach that can be applied to solve the clustering problem is the use of metaheuristics, which are procedures capable of escaping from local optima, once the use of exact methods is computationally infeasible. However, the majority of the metaheurísticas applied to clustering problem is not scalable for real or commercial bases. They are more effective for smaller instances of the problem trated. The computational cost necessary to calculate the solutions becomes greater in larger instances of the problem. For this reason, hybrid procedures that explore the combination of metaheuristics represent a promising approach for solving the clustering problem. This work shows a proposal of a Hybrid Genetic Clustering Algorithm that associates the process of global search to a local search heuristic and also initializes the population by different grouping techniques. Such improvements aim to direct the search for solutions next to the global optimal one. An experimental evaluation with real and synthetic databases is performed aiming to verify if the proposed approach presents an improvement in relation to the other evaluated algorithms. The result of this analysis shows that the proposed algorithm presents a better performance in four among the six evaluated algorithms. In addition, an analysis of the execution time shows that the execution time of our proposal is feasible, even though it is considerably longer than the execution times of the fast convergence algorithms.