Detection of distributed denial of service attacks at source

• Autor(a) principal: Silva, Fábio Alexandre Henriques da
• Data de Publicação: 2018
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/27819
• topic: Patterns recognition; Dynamics analyzes; Network activity modeling; Methodologies for anomalies detection; Network activity classification

From year to year new records of the amount of traffic in an attack are established, which demonstrate not only the constant presence of distributed denialof-service attacks, but also its evolution, demarcating itself from the other network threats. The increasing importance of resource availability alongside the security debate on network devices and infrastructures is continuous, given the preponderant role in both the home and corporate domains. In the face of the constant threat, the latest network security systems have been applying pattern recognition techniques to infer, detect, and react more quickly and assertively. This dissertation proposes methodologies to infer network activities patterns, based on their traffic: follows a behavior previously defined as normal, or if there are deviations that raise suspicions about the normality of the action in the network. It seems that the future of network defense systems continues in this direction, not only by increasing amount of traffic, but also by the diversity of actions, services and entities that reflect different patterns, thus contributing to the detection of anomalous activities on the network. The methodologies propose the collection of metadata, up to the transport layer of the osi model, which will then be processed by the machien learning algorithms in order to classify the underlying action. Intending to contribute beyond denial-of-service attacks and the network domain, the methodologies were described in a generic way, in order to be applied in other scenarios of greater or less complexity. The third chapter presents a proof of concept with attack vectors that marked the history and a few evaluation metrics that allows to compare the different classifiers as to their success rate, given the various activities in the network and inherent dynamics. The various tests show flexibility, speed and accuracy of the various classification algorithms, setting the bar between 90 and 99 percent.