An Efficient Framework for Unsupervised Anomaly Detection over Edge-Assisted Internet of Things

The detection of anomaly status plays a pivotal role in the maintenance public transportation and facilities smart cities. Owing to pervasively deployed sensing devices, one can collect apply multi-dimensional data detect analyze potential anomalies react promptly. Current efforts concentrate on offline manners fail fit situation cities, where efficient online solutions are expected. In this paper, novel framework is designed for over edge-assisted Internet-of-Things (IoTs). allows periodical collection from sensors continuous at edge node. A unsupervised deep learning model balance resource consumption accuracy detection, based combination convolutional autoencoder adversarial training. Meanwhile, proposed also adopts an adaptive strategy reduce overall consumption. According theoretical analysis evaluation several real-world datasets, discover correlation features among data, efficiently abnormality