Minimal but Lethal: A XAI-Driven Approach for Feature-Level Adversarial Attacks on Healthcare 5.1

In Healthcare 5.0, the expanded attack surface increases the vulnerability of Intrusion Detection Systems (IDS) to sophisticated threats. Among them, adversarial attacks modify features to evade the detection of malicious samples. XAI-driven methods enable the manipulation of fewer — sometimes just one—features while maximizing impact. To date, no XAI-driven adversarial strategy has been applied to cyber-biomedical features in Healthcare 5.0. In this work, we address this gap by employing XAI-Driven approach to maximize IDS degradation through a feature-level adversarial attacks. Our results reveals that a single feature perturbed can drastically reducing F1-Score from 99% to 0% in data alteration scenarios and from 81% to 12% in spoofing attacks.