Adaptive Hierarchical GHSOM with Federated Learning for Context-Aware Anomaly Detection in IoT Networks

This paper proposes an adaptive hierarchical Growing Hierarchical Self-Organizing Map (GHSOM) integrated with Federated Learning (FL) for scalable and privacy-preserving anomaly detection in distributed Internet of Things (IoT) networks. In IoT environments, vast amounts of data are generated from heterogeneous sensors, requiring robust anomaly detection mechanisms to ensure system reliability. Traditional centralized approaches face significant challenges related to data privacy, communication overhead, and scalability. The proposed model addresses these issues by combining the local processing power of GHSOM with the privacy-preserving capabilities of FL, ensuring that only model updates, rather than raw data, are shared across the network. To further enhance efficiency, we introduce a novel adaptive hierarchy within the GHSOM structure. This hierarchy dynamically adjusts based on the context and characteristics of the incoming sensor data, such as its variance, frequency, and criticality. By embedding this context-awareness, the GHSOM model can tailor its clustering process to the specific requirements of each sensor node, improving anomaly detection accuracy. Additionally, FL is enhanced with a selective update mechanism, where only significant model updates—triggered by detected anomalies—are shared across the network, further reducing bandwidth usage and communication costs. We validate the proposed system on diverse IoT sensor datasets, demonstrating its superior ability to detect anomalies with reduced communication overhead compared to non-adaptive federated GHSOM approaches. This work highlights the potential of combining adaptive GHSOM hierarchies with federated learning to create scalable, privacy-preserving, and efficient anomaly detection systems suitable for real-time IoT applications.