EUVIMEDEUVIMED entdecken
Crossref

A Lifecycle-Oriented Explainability Framework for Trustworthy Decision-Making in Intelligent Computing Systems

George Zacharia

QP-AIDSE · 2026

DOI: 10.65713/qpv2i201 ↗

Kurz gefasst

Fachartikel in der Zeitschrift QP-AIDSE, 2026. Verfasst von George Zacharia. Die Arbeit nennt 25 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Artificial intelligence (AI) increasingly supports classification, prediction, ranking, and automated decision-making, yet complex models can provide limited insight into the evidence behind their outputs. This opacity complicates auditing, error analysis, fairness assessment, and appropriate human oversight. This paper proposes a lifecycle-oriented framework that integrates explainable artificial intelligence (XAI) with uncertainty assessment, robustness checks, fairness analysis, and governance. The framework separates predictive inference from explanation generation and decision assurance, allowing model outputs to be accompanied by local or global explanations, reliability indicators, and escalation rules. Established approaches, including interpretable models, Local Interpretable Model-Agnostic Explanations (LIME), Shapley Additive Explanations (SHAP), and permutation-based feature importance, are discussed in relation to fidelity, stability, computational cost, and human usability. A reproducible benchmark experiment is conducted on the public Wine Recognition dataset using logistic regression and random forest classifiers. The data are partitioned into training and held-out test sets with stratification, and model performance is evaluated using accuracy, macro-precision, macro-recall, macro-F1, and log loss. Permutation importance is computed on the held-out set to demonstrate a model-agnostic explanation workflow. The experiment is intentionally presented as a limited methodological demonstration rather than evidence of universal superiority or deployment readiness. Results show the measured behaviour of the two models on this benchmark and illustrate how predictive metrics and feature-attribution evidence can be reported together. The paper concludes that explanations should not be treated as proof of correctness: trustworthy decision-making requires explanation validation, uncertainty communication, domain-appropriate fairness and robustness tests, and lifecycle monitoring.

Zitieren

APA

Zacharia, G. (2026). A Lifecycle-Oriented Explainability Framework for Trustworthy Decision-Making in Intelligent Computing Systems. QP-AIDSE. https://doi.org/10.65713/qpv2i201

Vancouver

Zacharia G. A Lifecycle-Oriented Explainability Framework for Trustworthy Decision-Making in Intelligent Computing Systems. QP-AIDSE. 2026. doi:10.65713/qpv2i201

BibTeX
@article{Zacharia2026,
  author = {Zacharia, George},
  title = {A Lifecycle-Oriented Explainability Framework for Trustworthy Decision-Making in Intelligent Computing Systems},
  journal = {QP-AIDSE},
  year = {2026},
  doi = {10.65713/qpv2i201}
}
Kostenlos starten →