| Explainable Artificial Intelligence (XAI) is a fundamental requirement in security-sensitive domains, where automated inferences must remain transparent, auditable, and subject to human oversight. This paper presents a transparency-by-design framework for explainable text classification in security and risk monitoring, integrating lexically traceable textual representations, an inherently interpretable linear model, and a structured explanation layer providing both global and local attribution insights. The framework is evaluated in a real-world conflict monitoring setting based on narrative event reports, achieving 94\% accuracy while preserving full traceability of the decision process. Attribution analysis shows that predictions are driven by semantically meaningful linguistic indicators, with consistent alignment between model outputs and human-interpretable features. The results demonstrate that transparency and predictive performance can be jointly achieved, supporting the development of trustworthy and auditable AI systems for security event analytics. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.