Deep reinforcement learning based intrusion detection system with feature selections method and optimal hyper-parameter in IoT environment

Document Type

Conference Proceeding

Publication Title

Proceedings of the 2022 International Conference on Computer, Information and Telecommunication Systems, CITS 2022

Publisher

IEEE

School

School of Engineering

RAS ID

45471

Funders

Beijing Natural Science Foundation (No. 4212015) / China Ministry of Education - China Mobile Scientific Research Foundation (No. MCM20200102) / Deanship of Scientific Research at King Khalid University for funding this work through Large Groups Project under grant number RGP.2/201/43

Comments

Bakhshad, S., Ponnusamy, V., Annur, R., Waqasyz, M., Alasmary, H., & Tux, S. (2022, July). Deep reinforcement learning based intrusion detection system with feature selections method and optimal hyper-parameter in IoT environment. In 2022 International Conference on Computer, Information and Telecommunication Systems (CITS) (pp. 1-7). IEEE. https://doi.org/10.1109/CITS55221.2022.9832976

Abstract

The continuous rise of inter-contented Internet of Things (IoT) devices has significantly increased network traffic, complexity, and the ever-changing Internet environment, making them more vulnerable to security attacks. Therefore, a robust and elegant intrusion detection system (IDS) based on advanced machine learning methods is required for securing the IoT environment. This paper discusses the new deep reinforcement learning (DRL) based network intrusion detection system (NIDS) with feature selection methods. However, the structure and training of the DRL model are still challenging tasks. Moreover, the effectiveness and accuracy of DRL-IDS crucially depend on the suitable hyper-parameters adaptation, i.e., differing hyperparameters can result in markedly varied IDS performance. Furthermore, due to the commercial value of hyper-parameters, confidentiality may be deemed necessary, and proprietary algorithms may protect their exclusive use. Therefore, we find different optimal hyper-parameters values for the training of DRL agents. Furthermore, we evaluate the effectiveness of different hyper-parameters both theoretically and empirically. For instance, we assess the hyper-parameters for the case of varying routing systems and countermeasures and integrate the optimal hyper-parameters for various network performances.

DOI

10.1109/CITS55221.2022.9832976

Access Rights

subscription content

Link to Full Text

Share

 
COinS