AI-Based Network Intrusion Detection System for Enterprise and Cloud-based Infrastructures

Abstract

The proliferation of enterprise networks, distributed cloud structures, and the Internet of Things (IoT) has increased the dangers and vulnerabilities of digital infrastructure to sophisticated cyberattacks. Traditional Network Intrusion Detection Systems (NIDS) rely solely on static signature matching, which can't detect new zero-day attacks, and standalone systems that use anomaly detection techniques suffer from high false-positive rates that can overload Security Operations Centres (SOCs). One of the main problems is accurately identifying threats while maintaining the detection system's adaptability and scalability, which challenge is increasingly addressed by leveraging Artificial Intelligence (AI) and deep learning. The literature review and synthesis are systematic and conducted using peer-reviewed literature on the topic from 2018 to 2026, following an explicit methodology aligned with the PRISMA guidelines. The takeaway messages are that cost-sensitive and deep hybrid architectures outperform both the cost-sensitive and cost-ignorant benchmarks, but their effectiveness critically depends on model explainability, evaluation rigour, and threshold tuning. Lastly, this survey summarizes some of the biggest challenges and future research directions in explainable AI (XAI), federated learning, as well as lightweight and resource-aware modeling for network security.