Uma nova proposta de paralelismo e balanceamento de carga para o algoritmo Apriori

• Autor(a) principal: Bolina, André Camilo
• Data de Publicação: 2013
• Outros Autores: Pereira, Denilson Alves, Esmin, Ahmed Ali Abdalla, Pereira, Marluce Rodrigues
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Repositório Institucional da UFLA
• Texto Completo: http://repositorio.ufla.br/jspui/handle/1/11504
• topic: Algoritmos computacionais; Mineração de dados; Balanceamento de carga; Computer algorithms; Data Mining; Load balancing; Distributed Parallel Apriori (DPA); Distributed Multithread Apriori (DMTA)

The main goal of data mining is to discover relevant information on digital content. The Apriori algorithm is widely used to this objective, but its sequential version has a low performance when executed over large volumes of data. Among the solutions for this problem is the parallel implementation of the algorithm, and among the parallel implementations presented in the literature that based on Apriori, it highlights the DPA (Distributed Parallel Apriori). This paper presents the DMTA (Distributed Multithread Apriori) algorithm, which is based on DPA and exploits the parallelism level of threads in order to increase the performance. Besides, DMTA can be executed over heterogeneous hardware platform, using different number of cores. The results showed that DMTA outperforms DPA, presents load balance among processes and threads, and it is effective in current multicore architectures.