A Deep Reinforcement Learning Approach to Edge-based IDS Packets Sampling

Publisher:
Institute of Electrical and Electronics Engineers (IEEE)
Publication Type:
Conference Proceeding
Citation:
2022 5th International Conference on Data Science and Information Technology, DSIT 2022 - Proceedings, 2022, 00, pp. 1-6
Issue Date:
2022-01-01
Filename Description Size
A_Deep_Reinforcement_Learning_Approach_to_Edge-based_IDS_Packets_Sampling.pdfPublished version450.14 kB
Adobe PDF
Full metadata record
Edge computing expands the Internet of Things (IoT) by allowing partial computing tasks to be migrated to edge servers and alleviate the computing pressure of terminal devices. However, the edge servers will be attacked through malicious network traffic. A well-designed edge-based IDS plays a significant role in protecting from malicious attacks. In this paper, we employ a gated recurrent unit classifier to perform intrusion detection due to its characteristics on long-term memory of inputs and light structure. Furthermore, to reduce the cost of performing intrusion detection when facing a large volume of data, we propose an actor-critic network based on deep reinforcement learning for packets sampling to work on some packets by ignoring the others. The system we designed can achieve great classification performance through partial packets. We use dataset CIC-IDS-2017 to evaluate our model, and the accuracy of classification reaches 97% while the proportion of detection packets is 26% and below. Evaluations of our approach on UNSW-NB15 and CIC-IDS-2017 show that our model maintains under 26% and 18% of the selected packets, respectively.
Please use this identifier to cite or link to this item: