Traffic-Aware Conformal Prediction for Robust Network Intrusion Detection

To prevent conformal prediction thresholds from failing under adversarial perturbation, this method calibrates on traffic drawn from expected attacker mechanisms. By excluding attacker-controllable features from the scored representation, the framework provides exact, pathwise coverage guarantees against adaptive attackers.