Abstract
An Optical Bust Switching (OBS) network is vulnerable to a range of issues. One of the most significant issues is Burst Header Packet (BHP) flooding attacks, which can negatively impact on the Quality of Service (QoS) and create more urgent issues such as Denial of Service (DoS). Existing techniques for countering BHP flood attacks usually display a low accuracy in detecting misbehaving nodes leading to BHP attacks. By contrast, Machine Learning (ML) is a widely adopted and powerful data analysis technique which has showed a high degree of predictive performance in multiple application domains due to its ability to discover beneficial knowledge for decision-making. This study investigates the use of predictive ML to counter the risk of BHP flooding attacks experienced in OBS networks, proposing a decision tree-based architecture as an appropriate solution. This contains a learning algorithm that extracts novel rules from tree models using data processed from several simulation runs. The results show that the rules derived from our learning algorithm will accurately classify 93% of the BHP flooding attacks into either Behaving (B) or Misbehaving (M) classes. Moreover, the rules can further classify the Misbehaving edge nodes into four sub-class labels with 87% accuracy, including: Misbehaving-Block (Block), Behaving-No Block (No Block), Misbehaving-No Block (M-No Block), and Misbehaving-Wait (M-Wait). The results clearly show that our proposed decision tree model is a viable solution compared to decisions undertaken by expert domains or human network administrators.