AI Classifiers Comparison for Network Anomaly Behavior Analysis

As the number of devices connected through mobile networks increase exponentially, the communication demands a broader range of services that can be provided by the fifth generation of communication networks (5G). However, with the increase of devices sharing data and services, the surface of attack also increases, leading to known and unknown threats that can affect the quality of the communications. An efficient way to detect threats in this scenario, is by analyzing the behavior of the data in the network. The use of machine learning algorithms for this type of application is on the rise, as they are an efficient way for detecting and classifying anomalies. In this work, three different machine learning techniques are tested to detect cyberattacks targeting the integrity of the communication where Botnet, DoS and infiltration attacks were launched. Additionally, two dimensionality reduction techniques were compared to evaluate the performance of the AI techniques under constrained information scenarios. Results show that the selection of the machine learning technique is crucial to obtain better results for given attack scenarios and different dimensions.