FedGroup: A Federated Learning Approach for Anomaly Detection in IoT Environments

The increasing adoption and use of IoT devices in smart home environments have raised concerns around the data security or privacy of smart home users. Several studies employed traditional machine learning to address the key security challenge, namely anomaly detection in IoT devices. Such models, however, require transmitting sensitive IoT data to a central model for training and validation which introduces security and performance concerns. In this paper, we propose a federated learning approach for detecting anomalies in IoT devices. We present our FedGroup model and algorithms that train and validate local models based on data from a group of IoT devices. FedGroup also updates the learning of the central model based on the learning changes that result from each group of IoT devices, rather than computing the average learning of each device. Our empirical evaluation of the real IoT dataset demonstrates the capability of our FedGroup model and anomaly detection accuracy as the same or better than federated and non-federated learning models. FedGroup is also more secure and performs well given all the IoT data are used to train and update the models locally.