A Domain-Invariant Multi-Scale Graph Temporal Learning for Cross-Environment Wi-Fi Deauthentication Detection

The study introduces GraphBurst-InvarNet, a leakage-resistant framework that combines dual-scale temporal learning, dynamic AP–client graph modelling, adaptive structural gating, and domain-adversarial training using hybrid dataset. Its main contribution is improved cross-environment detection of sparse-to-flood Wi-Fi deauthentication attacks, achieving 97.85% external Macro-F1 with only 0.38% false positives on an unseen dataset.