Advanced Machine Learning Approaches for Zero-Day Attack Detection: A Review

Zero-day attacks provide an essential challenge in cybersecurity because of their unpredictability and absence of pre-existing defenses. To detect these threats, this paper thor-oughly examines machine learning (ML) and artificial intelligence (AI) methodologies, encompassing supervised, unsupervised, and hybrid models. It underscores the capabilities of modern AI technologies, including deep learning, federated learning, and lightweight AI models, especially in real-time detection and resource-constrained environments. The research highlights the considerable deficiencies in the availability and uniformity of zero-day datasets, discusses the advantages and limitations of ML-based detection methods, and proposes directions for future inquiry, such as adversarial learning, privacy-preserving strategies, and the enhancement of real-time detection. The results intend to assist researchers and practitioners in formulating more resilient, scalable approaches to address zero-day vulnerabilities.