Adaptive Weight XGBoost: Detecting and Mitigating Low-Rate DoS Attack in Network Slicing

Network slicing is an emerging architecture that allows multiple virtual networks to be created on top of a shared physical infrastructure, each tailored to a specific type of service or application. Managing network slices using a Software Defined Network (SDN) controller makes the network more scalable and manageable but also vulnerable due to the centralized control of SDN. Low-rate denial-of-service (LDoS) is periodic and stealthy, and its attack frequency is lower than that of ordinary distributed denial-of-service attacks, making it one of the most severe threats on SDN. To cope with the above challenges, we propose a real-time Lightweight LDoS Detection and Mitigation (L
²
DM) framework. Moreover, an Adaptive Weight XGBoost (AW-XGBoost) algorithm is designed to extract features and obtain the detection model via adjusting adaptive weights. We also leverage the architecture of network slicing to build honeypot slices to gather information from attackers. Experimental results show that the proposed framework and corresponding algorithm can be deployed on SDN controllers to achieve LDoS attack detection and mitigation at low cost with high accuracy and effectiveness.