Ontology-Enhanced Neuro-Symbolic Defense against Adversarial Attacks in IoT-Enabled Cyber-Physical Systems: Experimental Validation in Smart Campus Infrastructure

Abstract

The rapid growth of Internet-of-Things (IoT) technologies has significantly contributed to the evolution of Cyber-Physical Systems (CPS), particularly within smart campus infrastructures. Despite these advancements, IoT-based CPS are increasingly vulnerable to adversarial attacks that can compromise data integrity, sensor accuracy, and system safety. Traditional AI-based defense mechanisms often lack contextual awareness and interpretability. This paper introduces an ontology-enhanced neuro-symbolic framework to detect and mitigate adversarial attacks in IoT-enabled CPS environments. Our approach integrates domain ontologies with a hybrid architecture that combines symbolic reasoning and deep learning to enhance resilience and explainability. The ontology captures semantic relationships among entities such as sensors, data streams, physical contexts, and network behavior within the smart campus ecosystem. The neuro-symbolic engine processes this structured knowledge alongside raw sensor data, enabling context-aware anomaly detection and response. To validate the proposed system, we deploy it in a real-world smart campus testbed comprising over 150 IoT nodes, including surveillance cameras, HVAC controllers, environmental sensors, and access control units. The system is tested against a range of adversarial attacks including data poisoning, model evasion, and logic manipulation. Experimental results demonstrate a 27% increase in adversarial detection accuracy compared to standard CNN and RNN models, with a 19% improvement in false-positive reduction. Furthermore, symbolic inference allows for better interpretation of attack sources and propagation paths. The fusion of ontological context and machine learning outputs leads to actionable insights for campus administrators and security personnel. This study underscores the importance of semantic knowledge in improving AI robustness and sets the groundwork for scalable, interpretable, and resilient defense systems in CPS. Future work will explore extending the ontology to inter-campus networks and integrating federated learning to ensure privacy-preserving collaboration.