Intrusion Detection for Internet of things Using Self-supervised Pre-training on GRU
Fujun Pei, Yuke Chang, Mingjie Shi · 2022
Intrusion detection is crucial in the Internet of Things (IoT) due to the scarcity of computing resources and the variability of the network environment when it is used in smart buildings, smart factories and other scenarios. Current intrusion detection solutions are mostly applicable to a single environment (or a single dataset) and require large amounts of labeled data for supervised training methods, which are not applicable to variable IoT network environments. In this paper, we investigate the recognition effect of GRU networks on the packet sequences made up of IoT network packets and suggest a self-supervised pre-training method based on GRU networks. In order to improve the effectiveness of supervised training in the following fine-tuning phase, the unlabeled data is utilized in the pre-training of this method to find sequence data correlations in the data flow and deeper representations of the data. The proposed model is evaluated on datasets such as TON-IoT, and it is observed that our method can significantly increase the model detection accuracy.