FedIoT-AD: Federated Anomaly Detection for Heterogeneous IoT Edge Networks

The rapid proliferation of Internet of Things (IoT) devices in smart home, industrial, and enterprise environments has introduced new challenges for anomaly detection. Centralizing data for model training introduces unacceptable privacy risks and communication overhead. This paper proposes FedIoT-AD, a federated learning framework for privacy-preserving anomaly detection across heterogeneous IoT edge networks. FedIoT-AD enables heterogeneous single-board computers (including OrangePi, Raspberry Pi, and BananaPi CM4) to collaboratively train a shared global anomaly detection model without transferring raw data to a central server. We address three core challenges: statistical heterogeneity (non-IID data distributions), system heterogeneity (varying compute constraints), and communication efficiency under bandwidth-limited edge conditions. Evaluation demonstrates anomaly detection performance within 2% of a centralized baseline while eliminating raw data transmission entirely and reducing communication overhead by 66% compared to standard federated averaging.