TriTrust: Risk-Aware Three-Factor Trust Fusion for Secure UAV Handover Admission
MATHEMATICS, cilt.14, sa.20, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 14 Sayı: 20
- Basım Tarihi: 2026
- Doi Numarası: 10.3390/math14203817
- Dergi Adı: MATHEMATICS
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, zbMATH, Directory of Open Access Journals, Academic Search Ultimate (EBSCO), Materials Science & Engineering Collection (ProQuest), Technology Collection (ProQuest)
- Kocaeli Üniversitesi Adresli: Evet
Özet
Cooperative unmanned aerial vehicle (UAV) swarms require secure transaction-level admission decisions when tasks, relay roles, control privileges, or mission assets are transferred between aircraft. Existing authentication and intrusion-detection mechanisms address complementary aspects of security and motivate integrating a mandatory protocol check with post-authentication behavioral and corroborative assessment. This paper introduces TriTrust, a risk-aware three-factor trust-fusion architecture that enforces a mandatory protocol-admission gate, followed by logistic fusion of traffic-behavior and competence-weighted witness scores for gate-passing candidates. The framework further incorporates validation-fitted score mapping and robust federated maintenance of the witness classifier. Evaluation on a constructed proxy-handover benchmark derived from public UAV cyber-physical records using cyber-condition-derived labels, sampled gate outcomes, and simulated witness views demonstrates that the modeled gate provides the dominant control over false acceptance: removing it increases mean seed-level raw-score false-acceptance rate (FAR) from 0.43% to 9.62%. Across five prespecified resamplings of the same source corpus, the isotonic-mapped full pipeline attains a mean seed-level area under the receiver operating characteristic curve (AUC) of 0.9926, a mean FAR of 0.83%, and a mean false-rejection rate (FRR) of 1.75%. Under 20% scaled sign-reversal poisoning, coordinatewise median aggregation maintains mean witness-classifier accuracy of 0.8552, compared with 0.2920 for sample-size-weighted averaging. Feature-space stress tests further show that false acceptance under learned-input manipulation remains limited by the modeled gate, whereas gate bypass combined with learned-factor compromise produces high false-acceptance rates. These results characterize the contributions and failure boundary of protocol gating, post-gate trust fusion, score mapping, and robust federated maintenance within the constructed benchmark.