Researchers from the University of Ottawa have introduced a new framework called SiamXBERT, which significantly improves the detection of unknown cyberattacks in Internet of Things (IoT) networks. This development comes at a time when the number of IoT devices is projected to rise from 19.8 billion in 2025 to about 31.2 billion by 2030, increasing security vulnerabilities.
Led by Shan Ali, Feifei Niu, Paria Shirani, and Lionel C. Briand, the framework effectively integrates flow-level and packet-level data, allowing it to perform well even with encrypted traffic. This is crucial as traditional detection systems struggle with the challenges posed by encryption and the dynamic nature of IoT communications.
SiamXBERT employs a meta-learning strategy to adapt quickly to new attack types using minimal training data. During tests on established IoT intrusion datasets, it achieved an impressive 78.8% improvement in identifying unknown threats compared to existing detection methods. This advancement highlights its potential for real-world applications in enhancing IoT security.
The researchers pointed out that traditional machine learning methods often fail against unknown threats, which include zero-day exploits. The rise in interconnected devices has contributed to an alarming increase in cyber threats, with over 1.5 billion IoT attacks reported by Kaspersky in just the first half of 2021.