Intrusion Detection Systems Utilizing Deep Learning: A Comprehensive Literature Review

Abstract

Certainly, with the increasing sophistication of cyberattacks, the need for intelligent Intrusion Detection Systems (IDSs) that can detect known and unknown threats in real time has become more critical than ever. Conventional IDSs (e.g., signature- and machine-learning based) are not effective against high-dimensional network traffic, zero-day attacks, and dynamic threat environments. With the ability to automatically capture complex features and learn hierarchical representations from a large volume of network data, Deep Learning (DL) has been a promising solution. In this review, the progress in deep learning IDS research from 2020 to 2026 is discussed. The study examines several architectures, such as Deep Neural Networks (DNNs), Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) networks, Autoencoders (AEs), Generative Adversarial Networks (GANs), and Transformer-based models. In addition, the commonly used benchmark datasets, performance measures, challenges, and future research directions are discussed. The review shows that the hybrid deep learning architecture consistently outperforms the traditional methods in terms of performance, with an accuracy of high detection performance and low false alarm. There are, however, open research challenges like dataset imbalance, model explainability, adversarial robustness and computational complexity. This paper utilizes keywords such as Deep Learning, CNN, LSTM, Transformer, Intrusion Detection System, Cybersecurity, Network Security, Anomaly Detection.