ELD: Adaptive Detection of Malicious Nodes under Mix-Energy-Depleting-Attacks Using Edge Learning in IoT Networks

Due to the distributed framework, Internet of Things (IoT) is vulnerable to insider attacks like energy-depleting attack, where an attacker can behave maliciously to consume the battery of IoT devices. Such attack is difficult to detect because the attacker may behave differently under various environments and it is hard to decide the attack path. In this work, we focus on this challenge, and consider an advanced energy-depleting attack, called mix-energy-depleting attack, which combines three typical attacks such as carousel attack, flooding attack and replay attack. Regarding the detection, we propose an approach called Edge Learning Detection (ELD), which can learn malicious traffic by constructing an intrusion edge and can identify malicious nodes by building an intrusion graph. To overcome the problem that it is impractical to provide labeled data for system training in advance, our proposed ELD can train its model during detection by labeling traffic automatically. Then the obtained detection results can be used to optimize the adaptability of ELD in detecting practical attacks. In the evaluation, as compared with some similar methods, ELD can overall provide a better detection rate ranged from 5% to 40% according to concrete conditions.