Energy-Efficient Adaptive Trust Self-Supervised Federated Graph Contrastive Learning for Intrusion Detection in Resource-Constrained VPN-IoT Networks

Abstract

A VPN-assisted Internet of Things (VPN-IoT) network is prone to different security threats owing to decentralized architectures, inadequate computational resources, and dynamic traffic patterns. The state-of-the-art intrusion detection techniques struggle to attain more detection accuracy while maintaining low energy consumption and preserving privacy. In order to handle these problems, this paper implements an Energy-Efficient Adaptive Trust Self-Supervised Federated Graph Contrastive Learning (EATSFGCL) framework. The proposed EATSFGCL integrates self-supervised contrastive learning and federated learning to discover significant traffic representations from unlabeled data while keeping raw network data on local devices. Graph neural networks are utilized in EATSFGCL to find the relationships among IoT devices and their communication activities where it assists to predict abnormal traffic patterns. An adaptive trust mechanism in EATSFGCL estimates the reliability of each mobile node and their model updates to decrease the influence of untrustworthy or attacked devices during federated training. In addition, an energy-aware optimization strategy in EATSFGCL considers device resource availability and communication costs to support efficient model training and deployment. Thus EATSFGCL improve attack detection performance under heterogeneous network conditions while minimizing energy usage and communication overhead. The effectiveness of EATSFGCL is determined in terms of detection accuracy, computational complexity, packet delivery ratio, energy consumption.