Recurrence Plots-Based Network Attack Classification Using CNN-Autoencoders

The advent of the Internet of Things, with the consequent changes in network architectures and communication dynamics, has affected the security market by introducing further complexity in traffic flow analysis, classification, and detection activities. Consequently, to face these emerging challenges, new empowered strategies are needed to effectively spot anomalous events within legitimate traffic and guarantee the success of early alerting facilities. However, such detection and classification strategies strongly depend on the right choice of employed features, which can be mined from individual or aggregated observations. Therefore, this work explores the theory of dynamic non-linear systems for effectively capturing and understanding the more expressive Internet traffic dynamics arranged as Recurrence Plots. To accomplish this, it leverages the abilities of Convolutional Autoencoders to derive meaningful features from the constructed plots. The achieved results, derived from a real dataset, demonstrate the effectiveness of the presented approach by also outperforming state-of-the-art classifiers.