AI-Driven Network Intrusion Detection Systems: A Systematic Review of Hybrid Models, Zero-Day Attack Mitigation, and Emerging Challenges in Cybersecurity
Sadiya Muhammad Rabiu, Bunyaminu Khalid Aminu, Dalhatu Aminu Zubairu · International Journal of Computer Applications · 2025
This systematic review synthesizes 45 peer-reviewed studies (2019-2024) on AI-driven Network Intrusion Detection Systems (NIDS) for enterprise cybersecurity.Advanced cyber threats, including zero-day exploits, adversarial AI, and ransomware, render traditional signature-based methods inadequate.AI-based NIDS, particularly hybrid models combining Machine Learning (ML) and Deep Learning (DL), exhibit superior detection accuracy, adaptability, and real-time responsiveness.Employing a PRISMA-guided methodology, this study evaluates hybrid ML-DL systems, zero-day detection techniques, adversarial countermeasures, and Explainable AI (XAI) frameworks.The meta-analysis indicates hybrid models achieve a mean accuracy of 96.2%, an F1-score of 0.94, and a 2.1% false positive rate, outperforming standalone ML (88.7% accuracy) and DL (92.5% accuracy) models by 10-15%.Real-world case studies in healthcare and smart cities, alongside cost-benefit analyses, demonstrate practical applicability.Standardized benchmarking protocols address dataset bias and adversarial vulnerabilities, validated in financial and healthcare sectors.The review proposes ethical AI frameworks, a future research roadmap, and deployment guidelines for enterprise Security Operations Centers (SOCs).