Causal-Contrastive Hypergraph Neural Network with Adaptive Trust for Zero-Day Attack Detection in Heterogeneous Wireless Sensor Networks

Abstract

Abstract— Heterogeneous Wireless Sensor Networks (HWSNs) are increasingly employed in environmental monitoring, industrial automation, healthcare, surveillance, smart agriculture, and intelligent infrastructure. The heterogeneous characteristics of sensor nodes provide flexibility in sensing and communication but also introduce significant cybersecurity challenges. Conventional intrusion detection systems generally depend on predefined attack signatures or supervised learning models trained using known attack categories and may have limited capability when previously unseen or coordinated attacks occur. This paper proposes a Causal-Contrastive Hypergraph Neural Network with Adaptive Trust (CCHG-AT) for zero-day attack detection in heterogeneous wireless sensor networks, extending the author's earlier work on lightweight elliptic-curve security for distributed systems [1] from communication-level protection to behavioural, learning-based detection. The framework constructs a dynamic hypergraph to represent high-order communication relationships, employs causal feature selection and contrastive representation learning, captures behavioural changes through temporal attention, and continuously evaluates node reliability through adaptive trust. The framework can be evaluated using suitable IoT/WSN intrusion-detection datasets and compared with machine-learning, deep-learning, graph, graph-attention, and hypergraph approaches.