A Survey of AI-Based Zero-Day Attack Detection Methods
Xueli Lin · Applied and Computational Engineering · 2025
Low disclosure rate and strong destructive force Zero-day attacks endanger people's property security, privacy rights, network service quality, and other interests. Although there are advanced artificial intelligence-based anti-zero-day-attack methods, there are a series of problems, such as the data sparsity problem, poor generalization ability. Although much research on AI zero-day attack detection since 2019 exists, there is still a lack of systematic study across multiple dimensions. In this paper, the research results of AI-based cybersecurity research on zero-day attack recognition from 2019 to 2025 are chronologically compiled, and how different cooperation mechanisms could best perform between the few-shot-learning method and the adversary-training method. Inspired by cybersecurity, deep learning with game theory, economic incentive structure mechanisms, and social behavior law ideas, this study proposes a new hybrid mode detection adaptivity system for cybersecurity based on the analysis. which can provide countermeasures and experimental suggestions to improve the effectiveness of various detection practices, and played an effective role in improving the intelligence level of the field of cybersecurity.