Enhancing Intrusion Detection Systems Using a Deep Learning and Data Augmentation Approach

Abstract

Cybersecurity relies heavily on the effectiveness of intrusion detection systems (IDSs) in securing business communication because they play a pivotal role as the first line of defense against malicious activities. Despite the wide application of machine learning methods for intrusion detection, they have certain limitations that might be effectively addressed by leveraging different deep learning architectures. Furthermore, the evaluation of the proposed models is often hindered by imbalanced datasets, limiting a comprehensive assessment of model efficacy. Hence, this study aims to address these challenges by employing data augmentation methods on four prominent datasets, the UNSW-NB15, 5G-NIDD, FLNET2023, and CIC-IDS-2017, to enhance the performance of several deep learning architectures for intrusion detection systems. The experimental results underscored the capability of a simple CNN-based architecture to achieve highly accurate network attack detection, while more complex architectures showed only marginal improvements in performance. The findings highlight how the proposed methods of deep learning-based intrusion detection can be seamlessly integrated into cybersecurity frameworks, enhancing the ability to detect and mitigate sophisticated network attacks. The outcomes of this study have shown that the intrusion detection models have achieved high accuracy (up to 91% for the augmented CIC-IDS-2017 dataset) and are strongly influenced by the quality and quantity of the dataset used.