Article: Feature selection and deep learning approach for anomaly network intrusion detection Journal: International Journal of Information and Computer Security (IJICS) 2024 Vol.23 No.4 pp.433 - 453 Abstract: The growth of internet-scale systems and data as well as increased cyber attacks call for more robust security solution. A crucial element in this regard is network intrusion detection system (NIDS), which analyses network traffic to detect attacks. However, a key challenge for NIDS is real-time detection of all possible attacks. The present work proposes a new approach, Chi2-GRU-MLP, to build an efficient anomaly network intrusion detection. With this innovative approach, we harness the capabilities of two deep learning models, namely gated recurrent unit and multi-layer perceptron, to make a suitable supervised classifier. Additionally, the uniqueness of this proposal lies in its utilisation of dependency feature selection that is used together with a proposed selection method for pre-processing the network traffic. This combination is designed to boost the efficiency and the acceleration of the proposed approach. The proposed approach is compared with other recent anomaly network intrusion detection works in terms of several metrics and using three different well-known datasets such as NSL-KDD, CICIDS-2017 and UNSW-NB15. Comparison shows that our proposed approach significantly outperforms earlier ANIDS. Inderscience Publishers - linking academia, business and industry through research

Title: Feature selection and deep learning approach for anomaly network intrusion detection

Authors: Khadidja Bennaceur; Zakaria Sahraoui; Mohamed Ahmad Nacer

Addresses: Department of Computer Science, Ecole Militaire Polythectnique, Bordj El Bahri, Algiers, Algeria ' Department of Computer Science, Ecole Militaire Polythectnique, Bordj El Bahri, Algiers, Algeria ' Department of Computer Science, Universite des Sciences et de Technologie Haouari Boumedien, Bab Ezzouar, Algiers, Algeria

Abstract: The growth of internet-scale systems and data as well as increased cyber attacks call for more robust security solution. A crucial element in this regard is network intrusion detection system (NIDS), which analyses network traffic to detect attacks. However, a key challenge for NIDS is real-time detection of all possible attacks. The present work proposes a new approach, Chi2-GRU-MLP, to build an efficient anomaly network intrusion detection. With this innovative approach, we harness the capabilities of two deep learning models, namely gated recurrent unit and multi-layer perceptron, to make a suitable supervised classifier. Additionally, the uniqueness of this proposal lies in its utilisation of dependency feature selection that is used together with a proposed selection method for pre-processing the network traffic. This combination is designed to boost the efficiency and the acceleration of the proposed approach. The proposed approach is compared with other recent anomaly network intrusion detection works in terms of several metrics and using three different well-known datasets such as NSL-KDD, CICIDS-2017 and UNSW-NB15. Comparison shows that our proposed approach significantly outperforms earlier ANIDS.

Keywords: anomaly network intrusion detection; deep learning; dependency-based feature selection; chi-square; gated recurrent unit; GRU; multi-layer perceptron; feature relevance degree.

DOI: 10.1504/IJICS.2024.139056

International Journal of Information and Computer Security, 2024 Vol.23 No.4, pp.433 - 453

Received: 24 May 2023
Accepted: 09 Jan 2024
Published online: 10 Jun 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: Feature selection and deep learning approach for anomaly network intrusion detection

Keep up-to-date