Machine Learning Based Defect-Detection Networks for Intrusion Detection, Using the NSL-KDD dataset: a comparative versus Deep learning Models

Abstract

Modern networks face an expanding and increasingly sophisticated set of security threats driven by the rapid growth of the Internet of Things (IoT), cloud computing, and high-speed communication technologies. Conventional, rule-based security systems struggle to keep pace with attacks that evolve in real time, motivating a shift toward intelligent, data-driven defect detection. This paper synthesizes the literature on Artificial Intelligence (AI), Machine Learning (ML), and Neural Networks (NNs) as applied to network security, tracing the evolution from early supervised classifiers to deep learning and, most recently, Explainable AI (XAI) and Self-Supervised Learning (SSL). It presents the principal categories of network threats, contrasts supervised and unsupervised learning paradigms, reviews the standard evaluation metrics used to judge detection performance, and consolidates twenty years of foundational and contemporary research into a structured comparative table and chart. The review concludes that while supervised methods deliver strong accuracy when labeled data is available, unsupervised and self-supervised approaches offer the scalability and adaptability required for real-world, label-scarce environments, and that interpretability remains an open research gap for AI-driven Defect Detection Networks (DDNs).