A Survey on the Applications of Deep Learning in Network Intrusion Detection Systems to Enhance Network Security

Network security breaches continue to increase in both complexity and impact, making intrusion detection a critical element of modern defense strategies. Intrusion Detection Systems (IDS) monitor system and network activity to identify malicious behavior and prevent potential damage. Among these, Network-based IDS (NIDS) play a particularly important role as they analyze network traffic at the packet or flow level, protecting diverse infrastructures. Recently, Deep Learning (DL) emerged as a transformative approach for NIDS by enabling the automatic learning of complex representations of normal and malicious traffic patterns without extensive manual feature engineering. This survey reviews recent advancements in DL-based NIDS and introduces a taxonomy comprising four categories: 1) reconstruction and generative models for anomaly detection; 2) transformer-based sequence models for capturing temporal dependencies; 3) convolutional and deep neural networks for supervised classification; and 4) hybrid and deployment-oriented approaches designed for real-world implementation. The review covers 31 studies published between 2020 and 2025, examining their datasets, feature sets, model architectures, and reported performance. A comparative analysis highlights prevalent practices and emerging trends. Finally, the survey identifies open challenges related to data quality, scalability, adversarial robustness, privacy, and interpretability, and outlines promising directions for future research. This work provides an updated synthesis of the field, a structured framework for comparative evaluation, and practical insights to guide the design of more secure and effective NIDS solutions.