Resumo

QUIALA FONSECA, Wilfredo. Data clustering from a parallel aproach. RCCI [online]. 2023, vol.17, n.4  Epub 01-Dez-2023. ISSN 2227-1899.

The DBSCAN clustering method is one of the best known density clustering methods due to its efficiency and simplicity. However, by its operation, it cannot address problems with a large number of samples where the execution time is considered relevant. At present, the grouping of large amounts of data is becoming an indispensable task. This problem is known as big data, where standard data mining techniques cannot cope with these data volumes. In this contribution, an approach based on parallelism with message exchange for DBSCAN clustering by density is proposed. This model allows us to classify a large number of unknown cases at the same time. For this, the mapping phase will determine the clusters in the different partitions of the data. Afterwards, the reduction phase will mix and update the clusters obtained from the previous phase. This model allows you to scale with data sets of arbitrary size, simply adding more compute nodes if necessary. In addition, this implementation obtains a clustering rate, similar to the clustering of the classical DBSCAN algorithm.

Palavras-chave : density clustering; clustering; parallel programming; DBSACN.

        · resumo em Espanhol     · texto em Espanhol     · Espanhol ( pdf )