IoT and Impacts to Data Breaches
Open Access DepositedA Hybrid Approach to Data Architecture Optimization
This novel study implements a compact, reproducible hybrid of machine learning and explainable AI to detect IoT-linked data-breach incidents in public narratives and to map model explanations to Zero Trust Architecture (ZTA) driver families. IoT involvement is labeled by a versioned distant-supervision lexicon, and lexicon-derived tokens are redacted from the narrative text branch prior to TF-IDF vectorization to support a leakage-controlled design. The corpus is divided into deterministic stratified training, validation, and test partitions. Logistic Regression, Random Forest, and XGBoost are tuned on validation data. A model-specific operating point, τ*, is selected on validation, locked, and evaluated once on the held-out test (80-10-10) split. Operational feasibility is determined under a fixed deployment policy threshold (τ_deploy). A multi-pronged assessment of discrimination, probability-quality diagnostics and deployment behavior is applied. Although Random Forest leads PR-AUC and F1 at τ*, XGBoost is selected as the deployment finalist under the decision contract because it sustains materially higher recall at τ_deploy at a feasible alert volume. Multi-seed sensitivity runs support the robustness of this conclusion. Explanations from tree-based SHAP values are aligned to ZTA driver families, but since modest in share, the mapping is interpreted as an audit-ready, governance-aligned decision-support layer for control-family review rather than as a prescriptive governance rule set.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.