Optimized Auto Separate Federated Graph Neural With Enhanced Well-Known Signature Trust-Based Routing Attacks Detection in Internet of Things

Abstract

The term Internet-of-Things (IoT) refers to the interconnection of things to a physical network that is equipped with sensors, software, and other devices to share information among themselves. The objective of IoT is to enable objects to be accessible and interconnected through the internet. Thus, security for IoT devices is a significant problem because devices linked with the IoT network are resource-constrained. Also, exchanging information among nodes using internet attacks or insecure internet is aimed at destroying IoT standing Routing Protocol (RPL). To address those challenges, an Optimized auto Separate Federated Graph neural with enhanced well-known Signature trust-based Routing Protocol attack detection method (OSFG-SRPL) is proposed. It undergoes three stages such as behavior generation, sequence prediction, and trust analysis. Initially, double-layer angle multi-kernel extreme learning analysis and skill Fick's law optimization algorithms are proposed for the feature extraction and feature selection purpose. The trust evaluation is performed in terms of investigating the device's traffic flow and detecting its behavior deviations in the attack environment, which is called a sequence prediction issue. It is efficiently handled by the proposed auto Separate Osprey Federated Graph neural network with Node-level capsule Bi-directional Long Short-Term Memory (SOFG-NBiLSTM) network. Finally, the introduced approach predicts the traffic behavior based on historical behavior and deviation analysis, which is used for malicious node detection in the RPL attack scenario. The detection accuracy of the introduced system is 99.99% and 99.98% for the benchmark datasets RPL-NIDDS17 and RADAR, respectively, which is more efficient than the other methods.