A novel RPL defense mechanism based on trust and deep learning for internet of things

Along with the significant growth of applications and facilities provided by the Internet of Things (IoT) in recent years, security challenges and related issues to privacy become considerable interest of researchers. On the other hand, the de facto IoT routing protocol for low-power and lossy networks called RPL is vulnerable to various types of routing attacks. Many researchers have investigated RPL security solutions focusing on effective detection of prevalent and destructive routing attacks such as blackhole attack, selective forwarding attack, rank attack and so on. Recent studies are proposing trust-based mechanisms with the aim of replacing traditional cryptography-based operations with lightweight security models in order to cover the inherent challenges of IoT devices, including energy and computational limitations. Therefore, in this paper, focusing on the problem of RPL vulnerability against well-known routing attacks, we have proposed a trust-based attack detection model, which investigates traffic behavior in different attack scenarios and detects malicious nodes relying on behavior deviation exactly at the same time as the start of any attack activity. Expected behavior is predicted by our learning model trained from the historical routing behavior pattern, using recurrent neural networks as a powerful deep learning method, which leads to attack detection with high-level accuracy and precision. Both mathematical analysis and simulation results on multiple RPL attack scenarios show clearly that the proposed trust-based defense mechanism is an effective approach capable of timely and precisely detection of routing behavior pattern deviation of malicious nodes exactly at the start time of the attack occurrence, which leads to attack detection and attacker identification based on trust scores extracted from the detected fluctuations between expected and real routing behavior patterns.