Federated Threat Detection for Smart Home IoT rules

Guangjing Wang, Qiben Yan

Abstract: Smart homes are enhanced with the convenience offered by Internet of Things (IoT) devices. However, the interconnected behaviors of devices can lead to unexpected interactions, commonly referred to as interactive threats. This paper addresses the analysis of potential interactive threats in the IoT domain and introduces FedINT, a federated IoT interactive threat detection system. Building upon previous research, we represent device interactions in smart homes as interactive graphs. These graphs are then encoded using graph neural networks (GNNs). Considering the privacy concerns associated with smart home data, which is closely tied to users' daily lives, we propose a layer-wise clustering-based federated GNN framework. This framework allows collaborative training of the GNN model without sharing raw data and addresses statistical heterogeneity and concept drift issues specific to graph data. To evaluate our approach, we employ datasets collected from five IoT automation platforms. The results show that our prototype, FedINT, achieves an average accuracy exceeding 90% in detecting interactive threats, surpassing the performance of existing methods.