Deep Learning Based Network Intrusion Detection System for Resource-Constrained Environments

Rizvi, Syed; Scanlon, Mark; McGibney, Jimmy; Sheppard, John

Publication Date:  November 2022

Publication Name:  The 13th EAI International Conference on Digital Forensics and Cyber Crime

Abstract:   Network intrusion detection systems (IDS) examine network packets and alert system administrators and investigators to low-level security violations. In large networks, these reports become unmanageable. To create a flexible and effective intrusion detection systems for unpredictable attacks, there are several challenges to overcome. Much work has been done on the use of deep learning techniques in IDS; however, substantial computational resources and processing time are often required. In this paper, a 1D-Dilated Causal Neural Network (1D-DCNN) based IDS for binary classification is employed. The dilated convolution with a dilation rate of 2 is introduced to compensate the max pooling layer, preventing the information loss imposed by pooling and downsampling. The dilated convolution can also expand its receptive field to gather additional contextual data. To assess the efficacy of the suggested solution, experiments were conducted on two popular publicly available datasets, namely CIC-IDS2017 and CSE-CIC-IDS2018. Simulation outcomes show that the 1D-DCNN based method outperforms some existing deep learning approaches in terms of accuracy. The proposed model attained a high precision with malicious attack detection rate accuracy of 99.7\% for CIC-IDS2017 and 99.98\% for CSE-CIC-IDS2018.

Download Paper:

Download Paper as PDF

BibTeX Entry:


      @inproceedings{rizvi2022DLNIDS,
author={Rizvi, Syed and Scanlon, Mark and McGibney, Jimmy and Sheppard, John},
title="{Deep Learning Based Network Intrusion Detection System for Resource-Constrained Environments}",
booktitle="{The 13th EAI International Conference on Digital Forensics and Cyber Crime}",
series = {ICDF2C '22},
year=2022,
month=11,
location={Boston, USA},
publisher={Springer},
address = {New York, NY, USA},
abstract={Network intrusion detection systems (IDS) examine network packets and alert system administrators and investigators to low-level security violations. In large networks, these reports become unmanageable. To create a flexible and effective intrusion detection systems for unpredictable attacks, there are several challenges to overcome. Much work has been done on the use of deep learning techniques in IDS; however, substantial computational resources and processing time are often required. In this paper, a 1D-Dilated Causal Neural Network (1D-DCNN) based IDS for binary classification is employed. The dilated convolution with a dilation rate of 2 is introduced to compensate the max pooling layer, preventing the information loss imposed by pooling and downsampling. The dilated convolution can also expand its receptive field to gather additional contextual data. To assess the efficacy of the suggested solution, experiments were conducted on two popular publicly available datasets, namely CIC-IDS2017 and CSE-CIC-IDS2018. Simulation outcomes show that the 1D-DCNN based method outperforms some existing deep learning approaches in terms of accuracy. The proposed model attained a high precision with malicious attack detection rate accuracy of 99.7\% for CIC-IDS2017 and 99.98\% for CSE-CIC-IDS2018.}
}