A Quantitative Assessment of Cybersecurity Risk Scoring Models for Enterprise Database Systems in Cloud-Native Environments
DOI:
https://doi.org/10.63125/gk2krm47Keywords:
Cybersecurity, Risk Scoring, Cloud-Native, Databases, Vulnerability PrioritizationAbstract
Cybersecurity risk prioritization in cloud-native enterprise database environments has become increasingly complex because technical vulnerability severity alone does not adequately represent exploitation likelihood, asset importance, architectural exposure, or potential organizational consequences. This study quantitatively assessed and compared severity-based, exploitation-oriented, threat-informed, contextual, statistical, machine-learning, attack-path, and composite cybersecurity risk-scoring models for enterprise database systems. The final analytical dataset contained 48,216 vulnerability–asset observations retained from 52,684 initial records, with 98.7% data completeness, covering relational, NoSQL, distributed, database-as-a-service, and containerized database environments. High or critical CVSS vulnerabilities represented 58.5% of observations, whereas verified exploitation occurred in only 14.4%, demonstrating substantial divergence between technical severity and observed exploitation. CVSS-based prioritization achieved a ROC-AUC of 0.716 and PR-AUC of 0.341, compared with 0.846 and 0.573 for EPSS and 0.928 and 0.768 for attack-path scoring. The integrated composite model achieved the strongest overall performance, with 92.6% accuracy, 78.9% precision, 86.7% recall, 82.6% F1-score, 0.963 ROC-AUC, and 0.887 PR-AUC. Statistical analysis showed strong relationships between composite risk and attack-path probability (r = .66), EPSS probability (r = .61), and threat activity (r = .57). High EPSS probability increased exploitation odds by 3.42 times, while high-probability attack paths increased the odds by 3.17 times. Validation demonstrated stable performance, with composite ROC-AUC values of 0.959 under cross-validation, 0.957 in independent holdout testing, and 0.948 in temporal validation. Operationally, composite prioritization reduced priority remediation from 28,207 to 12,463 observations, representing a 55.8% workload reduction, while increasing high-risk coverage from 82.4% to 94.1%. The findings demonstrated that integrating exploitation likelihood, threat intelligence, asset context, privilege, exposure, security controls, and architectural dependencies provided a more accurate, robust, and operationally efficient basis for prioritizing cybersecurity risk in cloud-native enterprise database systems.


