Decision Reliability Evaluation of AI Expert Systems in High Impact Domains

This dataset is a supplementary research dataset for the manuscript entitled “Decision Reliability Evaluation of AI Expert Systems in High Impact Domains.” The dataset supports the analysis of decision reliability, interpretability consistency, and governance alignment in AI expert decision support systems applied to high-impact domains, including public services, healthcare decision support, and financial risk management. The dataset contains structured sheets covering dataset characteristics, baseline model performance, decision reliability statistics, interpretability consistency results, reconstructed 30-run experimental results, governance alignment mapping, model descriptions, and a codebook. The evaluated models include XGBoost, Deep Neural Network, and Random Forest. The main evaluation indicators include accuracy, precision, recall, F1-score, decision reliability score, confidence interval, standard deviation, p-value, and interpretability consistency based on SHAP attribution stability. This dataset was prepared to support transparency, reproducibility, and research documentation for the manuscript. The run-level values are provided as a reconstructed supplementary dataset based on the aggregate experimental results reported in the paper, rather than raw institutional records. Therefore, the dataset is suitable for academic documentation, validation of reported tables, secondary analysis, and repository submission purposes. The dataset is related to responsible AI evaluation, AI governance, repeated execution analysis, explainable AI, and high-impact decision support systems. It also supports the discussion of governance alignment with AI accountability principles, transparency requirements, SDG 9, and SDG 16.