Trust Based IOT Routing Attacks Detection Using Recurrent Neural Networks

Along with the recent growth of IOT applications, related security issues have also received a great attention. Various IOT vulnerabilities have been investigated so far, among which, internal attacks are the most important challenge that are mostly aimed at destroying IOT standard routing protocol (RPL). Recent studies have introduced trust concept as a practical tool for timely diagnosis and prevention of such attacks. In this paper trust evaluation is performed based on investigating the traffic flow of devices and detecting their behavior deviations in case of RPL attack scenarios, which is formulated as a sequence prediction problem and a new Trust-based RPL Attacks Detection (TRAD) algorithm is proposed using Recurrent Neural Networks (RNNs). Traffic behavior prediction based on historical behavior and deviation analysis, provides the possibility of anomaly detection, which has an enormous effect on the accuracy and predictability of attack detection algorithms. According to the results, the proposed model is capable of detecting compromised IOT nodes in different black-hole and selective-forwarding attack scenarios, just at the beginning time of the first attack, which provides the possibility of early detection and isolation of malicious nodes from the routing process.