Shrink AutoEncoder for Federated Learning-based IoT Anomaly Detection

Federated Learning (FL)-based anomaly detection is a promising framework for Internet of Things (IoT) security. Due to the scarcity of abnormal data, unsupervised deep learning neural network models, such as variations of AutoEncoder (AE), are considered effective solutions for anomaly detection in IoT devices. These models construct low-dimensional representations of input data that are utilized for classification. Nevertheless, given the enormous number of IoT devices, their intrinsic heterogeneity, and the distributed nature of the FL training process, the latent representation of the local data is distributed randomly. The determination of the global anomaly score is thus no longer accurate. To address this issue, this work provides an effective FL-based IoT anomaly detection framework with novel AutoEncoder models, namely Federated Shrink AutoEncoder (FedSAE). The proposed model forces normal data of IoT devices to nearly the origin. Thus, a universal or global anomaly score can be determined accurately for all IoT devices. The extensive experiments on the N-BaIoT dataset indicate that FedSAE may reduce the false detection rate by 1.84% compared with that of the AE-based FL frameworks for the IoT anomaly detection problem.