The widespread adoption of unified standards like CSA Matter drives IoT interoperability but creates an urgent demand for security mechanisms that are scalable, privacy-preserving, and protocol-aware. Traditional centralized intrusion detection systems (IDS) fail to meet these requirements due to high bandwidth costs, latency, and privacy concerns under strict regulations such as GDPR and HIPAA. In this work, we present FedMAT, the first CSA Matter-specific federated intrusion detection system (IDS) that combines CoAP/TLV-layer semantic packet analysis with privacy-preserving model aggregation. FedMAT introduces a Hierarchical Edge–HPC Grid architecture that decouples the computational load across system tiers: Lightweight-matter endpoints perform low-latency inference with complexity \(\mathcal {O}(W \cdot L)\) , where L denotes the sequence length, while resource-intensive gradient computation with complexity \(\mathcal {O}(W \cdot L \cdot E)\) is offloaded to edge gateways (e.g., Border Routers). This hierarchical approach ensures feasibility on constrained hardware while leveraging differential privacy (DP) and secure multiparty computation (SMC) via the SecAgg protocol to provide rigorous data protection guarantees. The proposed system achieves 96.3% detection accuracy while reducing communication overhead by 74.2%, making it feasible for deployment in resource-constrained IoT environments. Real-world validation on a BSH smart appliance testbed and simulated cryptographic aggregation benchmarks with up to 10,000 virtual nodes demonstrate FedMAT’s scalability and its ability to handle millions of frames in congested wireless channels. Beyond current deployments, FedMAT provides a forward-looking architecture that can extend to cross-protocol IoT ecosystems and scale to thousands of devices in smart healthcare, industrial IoT, and smart city environments.
