Key points are not available for this paper at this time.
The rapid development of Internet of Things (IoT) accumulates lots of communication data. Not being processed in time, these massive data increase the difficulty of anomaly detection for smart service. Furthermore, labeling all communication data is unpractical. Therefore, it is necessary to accomplish some feasible ways which can incorporate unlabeled data effectively. In recent years, Temporal Convolutional Network (TCN) has been proposed to solve sequence problems and got better performance compared to Recurrent Neural Network (RNN) in most cases. This paper proposes a semi-supervised hierarchical stacking TCN for the first time, which concentrates on the anomaly detection of the communication in smart home. The ideology of hierarchical model fully considers the features of streaming data in smart home scenario, and the stacking method is used to weed out the outliers. In this way, the detection accuracy can be highly improved. Finally, experimental results demonstrate that the proposed model can promote the security of the communication in smart home with a large extent as well as get much better performance.
Cheng et al. (Tue,) studied this question.