Feature Engineering and Machine Learning Pipeline for Detecting Radio Protocol-based Attacks
Auwn Muhammad, Loay Abdelrazek, Ikram Ullah · 2023
Air interface is one of the most exposed interfaces in cellular networks facing the possibility of a variety of intentional and unintentional attacks that can be caused by rogue, compromised or misconfigured devices. Efficient real-time detection of such attacks is an indispensable pre-step for proper mitigation and response to minimize the impact of such attacks thus keeping RAN operations uninterrupted and ensuring smooth operations of devices using RAN services.In this paper, we take two control plane protocol-based attacks from earlier literature which aim at denial of service (DoS) over the air interface. Then we devise features used by machine learning algorithms and a generic pipeline for detecting such attacks under realistic traffic conditions. Our features, methods and pipeline are generic in terms of radio technology and attack detection, i.e., they are able to detect DoS attacks at different protocol layers and in different radio access technologies that uses similar control plane procedures. We show performance of our approach in terms of robustness of features and accuracy of detecting attacks targeting different layers of LTE RAN.