<![CDATA[AI expert decision support systems are increasingly used in public administration, healthcare, and financial risk management, yet conventional accuracy centered evaluations often fail to capture whether systems produce stable decisions across repeated executions. This study aims to develop a reliability oriented evaluation framework for assessing AI expert decision support systems beyond single-run predictive performance. The focus of the study is decision reliability in high-impact AI applications where inconsistent outputs may reduce accountability, weaken institutional trust, and create governance risks. A repeated experimental evaluation approach was applied using recent datasets from 2022 to 2024 representing heterogeneous and imbalanced decision conditions. The proposed framework integrates decision stability measurement, interpretability consistency assessment, confidence interval analysis, and statistical significance testing to examine system behavior under realistic operational scenarios. The results show that models with comparable predictive accuracy can demonstrate statistically significant differences in decision reliability. Confidence interval analysis indicates meaningful variability in output consistency, while interpretability evaluation reveals uneven explanatory stability across model executions. These findings confirm that reliability-oriented evaluation provides a more comprehensive and policy-relevant assessment of AI expert systems than accuracy-based evaluation alone. The study contributes to responsible AI deployment by offering an evaluation perspective that strengthens technical assessment, governance accountability, and trustworthiness in high-impact decision environments.]]>
Copyrights © 2026
