Hybrid AI-Based dynamic risk assessment framework with explainable AI practices for composite product cybersecurity certification
Гибридная система динамической оценки рисков на основе искусственного интеллекта с применением методов объяснимого ИИ для сертификации кибербезопасности составных продуктов
2026-02-10
SCID: 54.1/9h59mj6q
Discuss with AI
European Cybersecurity Certification SchemeExplainable AIcybersecurity certificationdynamic risk assessmentvulnerability exploitation prediction
Figures from the paper
Abstract (AI)
Abstract Cybersecurity certification generally relies on risk assessment results to identify suitable controls and assess the completeness of these controls for security requirement satisfaction and overall security assurance. Prioritization of relevant vulnerabilities is essential to support the risk assessment and overall conformity assessment. However, the security context has continuously evolved with variations in attack surfaces, vulnerability exploitation, and the regulatory landscape–factors that significantly impact the conformity assessment process. This research proposes a hybrid AI framework integrating ensemble learning with GPT-3.5 for effective risk management within composite product cybersecurity conformity assessment under the European Cybersecurity Certification Scheme. It operationalizes Explainable AI (XAI) practices using SHAP and LIME methods to identify the most influential features affecting vulnerability predictions, and applies marginal analysis to measure the quantifiable gap closure between required and actual security postures to validate security control adequacy and requirement satisfaction based on calculated risk levels. This facilitates the adoption of XAI in the context of cybersecurity certification, extending its utility beyond general AI-enabled application scenarios. An industrial pilot scenario based on the P-NET 5G/6G Testing and Integration Service infrastructure, along with a dataset-based experiment, was conducted to evaluate the proposed framework. The results indicate that the hybrid model achieved 89% accuracy for vulnerability exploitation score prediction, enabling accurate risk calculation for conformity assessment. Furthermore, the XAI analysis revealed that the identified security controls demonstrate adequate performance in satisfying mapped security functional requirements. Ultimately, the framework provides quantifiable validation of security control effectiveness, enabling auditors to trace the logical connections between vulnerability predictions, risk calculations, and security requirement satisfaction for an informed certification decision.
Key Findings
1
In a P-NET 5G/6G testing and integration infrastructure pilot and dataset experiment, the hybrid model achieved 89% accuracy in predicting vulnerability exploitation scores.
2
Marginal analysis quantifies gaps between required and actual security postures to validate control adequacy and mapped security-requirement satisfaction.
3
SHAP and LIME explainability methods identify influential features driving vulnerability-exploitation predictions, supporting transparent risk calculations and auditability.
4
The framework’s XAI analysis indicated that the evaluated security controls adequately satisfied their mapped security functional requirements and enabled quantifiable validation of control effectiveness.
5
The study proposes a hybrid cybersecurity risk-assessment framework combining ensemble learning and GPT-3.5 for composite-product certification under the European Cybersecurity Certification Scheme.
Research Object
composite products undergoing cybersecurity conformity assessment under the European Cybersecurity Certification Scheme, exemplified by the P-NET 5G/6G Testing and Integration Service infrastructure
Research Subject
dynamic cybersecurity risk assessment, vulnerability exploitation prediction, and explainable validation of security-control adequacy and security-requirement satisfaction
Publication Details
Publication Date
2026-02-10
Journal
Publisher
ISSN
Cited by
9
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest