Aprendizado de máquina para redução do tráfego de dados e da latência na névoa das coisas

Detalhes bibliográficos
Ano de defesa: 2020
Autor(a) principal: Alencar, Brenno de Mello
Orientador(a): Prazeres, Cássio Vinícius Serafim
Banca de defesa: Mendonça Neto, Manoel Gomes de, Delicato, Flavia Coimbra
Tipo de documento: Dissertação
Tipo de acesso: Acesso aberto
Idioma: por
Instituição de defesa: Universidade Federal da Bahia
Instituto de Matemática e Estatística
Programa de Pós-Graduação: em Ciência da Computação
Departamento: Não Informado pela instituição
País: brasil
Palavras-chave em Português:
Área do conhecimento CNPq:
Link de acesso: http://repositorio.ufba.br/ri/handle/ri/33634
Resumo: The Internet of Things (IoT) has produced infrastructures and applications that generate large amounts of data. These data are usually data streams, that have the characteristic of being continuous and infinite and also have the peculiarity of modifying their behavior over time. Due to the large capacity of storage, data processing, and provisioning of resources, this data is generally processed and analyzed in cloud computing environments. Although Cloud Computing provides the IoT infrastructure with adequate scalability and resource centric features, the distance between devices and the cloud can impose limitations to achieve low latency in data traffic. In order to maintain scalability, achieve low latency and reduce data traffic between the IoT devices and the Cloud, the Fog Computing was proposed. Although the Fog Computing paradigm establishes resource availability at the edge of the network, the technologies and techniques currently used for IoT data processing and analysis may not be sufficient to support the continuous and unlimited data stream that IoT platforms produce. In this way, this work presents an approach for processing and analyzing data stream from the Internet of Things in real time in Fog. The main advantage of using our approach is the possibility of reducing the amount of data transmitted on the network infrastructure, which allows, as a consequence, to perform an online data modeling, by detecting changes in data behavior, and a reduction of the Internet usage. In addition, the proposed platform does not require a constant Internet connection. Finally, we evaluate the proposal from the perspective of performance in a scenario of intelligent objects at the edge of the network.