Detecção de intrusão em nós sensores de redes de sensores sem fio

• Ano de defesa: 2021
• Autor(a) principal: Alves, Ian Nícolas Lemes
• Orientador(a): Machado, Renato Bobsin
• Banca de defesa: Franco, Edgar Manuel Carreno , Westphall, Carlos Becker
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual do Oeste do Paraná; Foz do Iguaçu
• Programa de Pós-Graduação: Programa de Pós-Graduação em Engenharia Elétrica e Computação
• Departamento: Centro de Engenharias e Ciências Exatas
• País: Brasil
• Palavras-chave em Português: Aprendizado de máquina; Redes de sensores sem fio; Nós sensores; Detecção de intrusão; Segurança computacional
• Palavras-chave em Inglês: Machine learning; Wireless sensor networks; Sensor nodes; Intrusion detection; Network security
• Área do conhecimento CNPq: SISTEMAS DE COMPUTACAO::TELEINFORMATICA
• Link de acesso: https://tede.unioeste.br/handle/tede/5905

Considering the wide usage of internet assisted computer systems and the sensitivity of their data, the application of systems and techniques that guarantee the security of that data becomes a necessity. The recent advancements in integrated technologies such as the Internet of Things, that are limited by their available resources, present challenges to the application of conventional security approaches like Intrusion Detection Systems, specially for Wireless Sensor Networks. This work aims to employ a security solution inspired by anomaly detection, applied to intrusion detection at the sensor node level in WSN. This is achieved by employing and comparing the Naïve Bayes, Multilayer Perceptron, Decision Tree and Random Forest algorithms, these comparisons are focused on the individual recall of each attack type, its average and, in terms of resource usage, the memory, energy and run time. The WSN-DS dataset, which deals with attacks to Wireless Sensor Networks, was used for training and testing and the proposed methodology is based on the Knowledge Discovery in Databases process, attribute selection was employed on the original dataset, analysis and tuning were performed to adjust algorithm parameters, according to the requirements of the evaluated sensor nodes. Through the algorithm execution and analysis of the performance metrics, the good performance of both the Decision Tree and the Random Forest algorithms could be observed, with the Random Forest algorithm displaying much higher resource consumption than the Decision Tree Algorithm.