A robust and machine learning-driven identification scheme for malicious nodes in UASNs

Abstract

Underwater Acoustic Sensor Networks (UASNs) have garnered widespread adoption across maritime domains and national security sectors. Their deployment, often in remote and challenging conditions, necessitates robust trust models to assess node integrity and to pinpoint compromised units, particularly under adversarial assault. Prevailing trust models, however, predominantly focus on nodes’ communicative actions and energy consumption, neglecting the influence of ambient aquatic noise on trust assessments. Moreover, these models typically falter amidst the UASN-specific impediments of high packet collision rates and elevated error probabilities inherent to underwater signal transmission, leading to consequential inaccuracies in node evaluation. To address these limitations, this paper proposes a robust and machine learning-driven malicious node identification scheme (RMIS) for UASNs. The proposed scheme first models and quantifies the impact of packet collisions and the underwater environment on node communication, utilizing existing environment models and MAC protocols. Second, communication traffic is exploited as effective and reliable evidence of trust. The traffic evidence can reflect multiple types of attacks by analyzing changes in traffic in a UASN. Third, the evaluation models are trained using support vector machine and K-means++ algorithms. Finally, two types of trust update mechanisms are proposed to handle the dynamic underwater environment and on–off attack. The simulation results indicate that compared to the other three identification schemes, RMIS is more effective in identifying malicious nodes. The effectiveness of the RMIS becomes particularly noticeable as the ratio of malicious nodes increases. Furthermore, the trust update mechanisms effectively limit the impact of the on–off attack and improve the robustness of the RMIS.