Um novo algoritmo imunológico artificial para agrupamento de dados
Arquivos
Tipo
Dissertação
Data de publicação
2010-01-27
Periódico
Citações (Scopus)
Autores
Borges, Ederson
Orientador
Silva, Leandro Nunes de Castro
Título da Revista
ISSN da Revista
Título de Volume
Membros da banca
Oliveira, Pedro Paulo Balbi de
Gomes, Lalinka Teixeira de Campos
Gomes, Lalinka Teixeira de Campos
Programa
Engenharia Elétrica
Resumo
Clustering is an important data mining task from the field of Knowledge Discovery in Databases. There are several algorithms capable of performing clustering tasks, and the most popular ones involve the calculation of a similarity or distance measure among objects from the database. Many algorithms can perform clustering in a simple and efficient manner, but have drawbacks as a way to get the optimal number of partitions and the possibility of getting stuck in local optima solutions. To try and reduce these drawbacks this dissertation proposes a new clustering algorithm based on Artificial Immune Systems. This algorithm is characterized by the generation of multiple simultaneous high quality solutions in terms of the number of partitions (clusters) for the database and the use of a cost function that explicitly evaluates the quality of partitions, minimizing the inconvenience of getting stuck in local optima. The algorithm was tested using four databases known in the literature and obtained satisfactory results in terms of the diversity of solutions, but has a high computational cost compared to other algorithms tested.
Descrição
Palavras-chave
agrupamento de dados , diversidade , k-médias , rede imunológica artificial , sistemas imunológicos artificiais , clustering , diversity , k-means , artificial immune network , artificial immune systems