Wireless Internet of Things (IoT) networks utilize Virtual Private Networks (VPN) to secure connections among heterogeneous and devices with limited computational resources. Though, encrypted VPN tunnels are still vulnerable to internal and compromised-node threats which cannot be significantly identified by state-of-the-art systems relying on packet payload analysis and centralized learning. With the aiming at handling these problems, this paper introduces a Hierarchical Spatio-Temporal Graph Auto-Encoder with Federated Contrastive Learning (HST-GA-FCL) for attack identification in VPN-supported wireless IoT environments. The HST-GA-FCL learns communication behavior with the assist of multi-level spatio-temporal graphs that capture device-, cluster-, and network-level relations. An unsupervised spatio-temporal graph auto-encoder in proposed HST-GA-FCL finds normal traffic patterns where a variation in reconstruction and latent representations is designated as attack. Federated learning in proposed HST-GA-FCL facilitates privacy-preserving collaborative training across IoT nodes while contrastive learning enhances representation reliability across assorted devices. The proposed HST-GA-FCL exactly predicts internal adversaries in encrypted VPN traffic with better accuracy. Simulation analysis across various IoT attack situations demonstrates that the proposed HSTGAE-FCL Model attains better performance in terms of detection accuracy, computational complexity, packet delivery ratio and false alarm rate when compared to existing systems.
Keywords: Attack, Federated Contrastive Learning, Internet of Things, Spatio-Temporal Graph Auto-Encoder, Virtual Private Networks
How to cite this article: Sangeetha S, Sudha L, Hierarchical Spatio-Temporal Graph Auto-encoded Federated Contrastive Learning For Attack Detection in VPN-Assisted Wireless IoT Network. Int J Drug Deliv Technol. 2026;16(3s): 136-144; DOI: 10.25258/ijddt.16.3s.18