Federated Learning Based Anamoly Detection for Secure and Resilient Internet of Things Communication Network

Abstract

The rapid proliferation of Internet of Things (IoT) devices has created highly distributed communication environments that are increasingly exposed to anomalous traffic, botnet activity, denial-of-service attacks, spoofing, malware propagation, and other cyber threats. Conventional centralized intrusion detection approaches require the collection and transmission of large volumes of potentially sensitive network data, creating privacy, communication overhead, scalability, and single-point-of-failure concerns. Federated Learning (FL) provides a distributed learning paradigm in which IoT nodes or edge clients collaboratively train a shared anomaly-detection model while retaining raw traffic data locally. This paper proposes a federated learning-based anomaly detection framework for improving the security, privacy, scalability, and resilience of IoT communication networks. The framework integrates local traffic analysis, feature extraction, distributed model training, privacy-preserving parameter exchange, and robust global aggregation to identify anomalous communication patterns without centralizing raw network observations. The proposed research considers heterogeneous and non-independent and identically distributed IoT traffic, resource-constrained devices, communication efficiency, model robustness, and adversarial threats against the federated learning process. Performance can be evaluated using accuracy, precision, recall, F1-score, false-positive rate, detection latency, communication cost, and convergence characteristics. Recent research confirms the growing relevance of FL-based intrusion and anomaly detection for IoT, while also identifying unresolved challenges involving heterogeneous data, aggregation security, resource constraints, and standardized evaluation.