Detecting Zero-Day Attacks in Internet of Medical Things Networks via Multi-Signal Open-Set Detection
Year: 2026
Type: conference-paper
Abstract: The main difficulty faced by traditional supervised intrusion detection systems is their unawareness of novel attack types the open-set problem. This paper proposes TRIDENT, a weighted sigmoid-based fusion approach combining three signals: Isolation Forest, SHAP Cosine Distance, and Deep Support Vector Data Description. TRIDENT identifies zero-day attacks in Internet of Medical Things (IoMT) networks by determining whether network flows genuinely belong to known classes. Fusion weights are optimized via grid search, and thresholds are calibrated to maximize F 2 -score, prioritizing recall for security-critical scenarios. Evaluated on two IoMT datasets CICIoMT2024 and IoMT-TrafficData, TRIDENT achieves AUC=89.4–100% and F 2 =0.789–1.000, outperforming the Isolation Forest baseline by +4.8–20.8%. Ablation studies confirm each signal’s contribution, and statistical tests (McNemar’s p<0.001) validate superiority over baselines. Robustness tests demonstrate graceful degradation under class imbalance and noise injection.
Source: Procedia Computer Science
Institutions: Mohamed I University, Université de Picardie Jules Verne
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7219180992
