Detecção preditiva de anomalias em redes de computadores com utilização de aprendizagem de máquina

Abstract

With the increasing dependence on technologies on a daily basis, it is evident the concern to maintain the infrastructures that support its operation, thus guaranteeing a good experience for the end user. Thus, denial of service attacks are among the main causes of anomalies in computer networks, which can cause degradation or even interruption of services. In this context, the application of new technologies, such as artificial intelligence or machine learning, becomes increasingly necessary to ensure more agility in detecting problems, reducing their impacts. Thus, this work presents an analysis between different methods of classifier supervised machine learning, applied to data collected fromnetwork equipment, of the switch type, in order to detect anomalies in the network infrastructure of a higher education institution. The machine learning methods used in this work were: Decision Tree, Random Forest, Extra Tree, Gradient Boosting, Extreme Gradient Boosting and Histogram Gradient Boosting. The models generated from these methods showed promise, being able to achieve results with 99.88% in the Weighted F1 metric and 99.16% of Balanced Accuracy. Other points, such as training time, prediction time and save file size, were also taken into account for the classification of the best method. Given the importance of fault detection tools, this work contributes to the definition of the best approaches and thus allows the development of new and more efficient tools for this purpose.