A systematic trust management system for RPL based IoT networks using machine learning

As IoT networks continue to evolve, concerns about their security risks are growing. Several security gaps still exist within IoT systems like Blackhole, Decreased Rank, Version Number, and Flooding attacks. During these attacks the intruder cheat legitimate nodes. Hence, affecting trustworthiness. One potential solution is the Trust Management System (TMS). TMS evaluates the trustworthiness of nodes, offers real-time predictions, detects attacks and aids decision-making to offer secure routing. This work proposes a lightweight, reliable and dynamic TMS to enhance IoT network security. There are two primary challenges in developing an AI-driven security system for IoT. The first is the availability of IoT datasets that include trust-based attacks and trust value as labels. The second is the inherent uncertainty of trust, which complicates the development of a dynamic system. The proposed TMS relies on three novel set of trust indicators: network flows, recommendations, and social behaviour. These indicators are aggregated using a modified Beta Distribution function to represent trust as labels in the dataset. The work also discusses the design of a lightweight and real-time trustworthiness predictor based on Kernel Extreme Learning Machine (KELM), which uses the labels from Beta Distribution for more accurate and reliable predictions. Furthermore, a dynamic threshold evaluation module is introduced to classify the nodes as either trustworthy or untrustworthy based on distance metrics. The proposed TMS is validated against four types of trust-based attacks-Blackhole, Decreased Rank, Version Number, and Flooding-which have not been adequately addressed in existing literature. The performance of proposed model is also validated on binary and multiclassification. Comparisons existing linear and non-linear machine learning models demonstrate that the proposed system outperforms current state-of-the-art TMS solutions, achieving 99.95% accuracy, 99.9% precision, 99.96% recall, and a minimal misclassification rate of just 0.05%.