Enhancing Network Security: the Role of Deep Learning and Explainable AI in Intrusion Detection Systems

NIDS stands for Network Intrusion Detection Systems; these are components of current integrated security systems that search for suspicious activity in the network traffic. Due to the growing scale and richness of network data, it becomes challenging to address such problems using traditional approaches based on machine learning. Various Deep Learning (DL) methods have been found to be efficient for identifying new and complex threats. However, the black box nature of these models is a major issue in high security risk environments where explainability and accountability are critical. This challenge is solved by Explainable AI (XAI) that offers understanding of how DL models are making decisions. This survey focuses on the combination of DL and XAI in NIDS and compares the advantages and disadvantages of different DL structures, such as CNNs, RNNs, LSTMs, and transformers and XAI methods such as SHAP, LIME, and attention. We focus on the issues of scalability, adversarial attacks, and real-time learning and identify further research opportunities in cross-modal learning, novel XAI techniques, and the emergence of self-learning, self-optimizing NIDS. The combination of DL and XAI could provide benefits to both approaches for NIDS, increasing the reliability of the intrusion detection systems.