Explainable AI-Driven Cyber Risk Analytics and Model Reliability Assessment for Intelligent Governance of U.S. Critical Infrastructure: An XGBoost and SHAP-Based Intrusion Detection Framework
Applied IT & Engineering · 2024
DOI: 10.25163/engineering.2110762 ↗Kurz gefasst
Fachartikel in der Zeitschrift Applied IT & Engineering, 2024. Die Arbeit wurde bisher 1-mal zitiert (Crossref).
Abstract (Originalsprache)
Background
U.S. critical infrastructure sectors — energy, healthcare, transportation, financial services, and communications — are increasingly governed by AI-driven digital technologies that, while operationally transformative, have dramatically widened the cyberattack surface. Distributed Denial of Service (DDoS) attacks, Advanced Persistent Threats (APTs), botnets, and ransomware now challenge infrastructure resilience in ways traditional, rule-based cybersecurity mechanisms were never designed to address. Despite machine learning's demonstrated promise for intrusion detection, most existing frameworks optimize narrowly for classification accuracy while neglecting the model reliability, explainability, and governance transparency that critical infrastructure operators actually require.
Methods
This study develops a resilient cyber risk analytics and model reliability assessment framework designed to support intelligent cybersecurity governance in U.S. critical infrastructure environments. Using the CICIDS2017 dataset — a realistic, labeled benchmark encompassing DDoS, brute force, botnet, web attack, and infiltration traffic — four supervised classifiers were trained and comparatively evaluated: XGBoost, Random Forest, Decision Tree, and Logistic Regression. Model performance was assessed across accuracy, precision, recall, F1-score, ROC-AUC, and false positive rate. SHAP (SHapley Additive Explanations) analysis was integrated to produce interpretable, feature-level explanations of model predictions, enabling governance actors to audit and act on classification outputs with informed confidence.
Results
The Support Vector Machine classifier achieved near-perfect discrimination on the binary DDoS versus benign classification task, with an AUC approaching 1.00, 2,571 true positives, 1,918 true negatives, and only eleven total misclassifications. Exploratory traffic analysis confirmed that flow duration, packet size, Flow Bytes/s, and destination port distribution carry substantial discriminative information for distinguishing attack from benign traffic. SHAP analysis identified the most influential network features driving model predictions, providing the feature-level transparency that governance decision-makers require.
Conclusion
Combining cyber risk analytics, machine learning, reliability evaluation, and explainable AI substantially advances cybersecurity resilience and governance trustworthiness for critical infrastructure protection — moving beyond detection accuracy toward systems that are interpretable, auditable, and operationally accountable.
Zitieren
APA
(2024). Explainable AI-Driven Cyber Risk Analytics and Model Reliability Assessment for Intelligent Governance of U.S. Critical Infrastructure: An XGBoost and SHAP-Based Intrusion Detection Framework. Applied IT & Engineering. https://doi.org/10.25163/engineering.2110762
Vancouver
Explainable AI-Driven Cyber Risk Analytics and Model Reliability Assessment for Intelligent Governance of U.S. Critical Infrastructure: An XGBoost and SHAP-Based Intrusion Detection Framework. Applied IT & Engineering. 2024. doi:10.25163/engineering.2110762
BibTeX
@article{euvimed2024,
title = {Explainable AI-Driven Cyber Risk Analytics and Model Reliability Assessment for Intelligent Governance of U.S. Critical Infrastructure: An XGBoost and SHAP-Based Intrusion Detection Framework},
journal = {Applied IT & Engineering},
year = {2024},
doi = {10.25163/engineering.2110762}
}
EUVIMED