كنفرانس بين المللی مهندسی صنايع و سيستم ها

صفحه اصلی / دوازدهمین كنفرانس بين المللی مهندسی صنايع و سيستم ها

A Trustworthy AI Governance Framework for AI-Assisted Project Evaluation

نویسندگان :

Behrouz Farhadzare¹ Zeinab Sazvar²

1- دانشگاه تهران 2- دانشگاه تهران

کلمات کلیدی :

Artificial Intelligence،Explainable Artificial Intelligence،Trustworthy AI،Project Evaluation،Project Governance،Decision Support Systems،Governance Readiness،Human Oversight

چکیده :

Artificial Intelligence (AI) is increasingly used to support strategic project evaluation by enhancing analytical capability and evidence-based decision making. However, organizations still face a critical challenge in determining whether AI-generated recommendations provide sufficient evidence for high-impact project decisions. Existing studies primarily emphasize predictive performance and model interpretability while providing limited guidance on governance mechanisms for responsible organizational adoption. This study proposes the Trustworthy AI Governance Framework (TAGF), which extends AI-assisted project evaluation from algorithm-centered prediction to governance-oriented decision support. TAGF integrates five complementary dimensions: predictive reliability, explainability evidence, explanation consistency, governance readiness, and human oversight. Three supervised machine learning models-Logistic Regression (LR), Decision Tree (DT), and Random Forest (RF)-were evaluated using repeated stratified cross-validation, conventional classification metrics, and SHapley Additive exPlanations (SHAP). Random Forest achieved the strongest predictive performance (Accuracy = 0.931, F1-score = 0.929, AUC = 0.972), followed by Decision Tree (Accuracy = 0.907, AUC = 0.951) and Logistic Regression (Accuracy = 0.889, AUC = 0.943). However, explanation consistency analysis revealed only moderate agreement among the models (global explanation correlations: −0.212 to 0.402), indicating that similar predictive performance does not necessarily imply similar analytical reasoning. The proposed framework reconceptualizes explainability as organizational governance evidence rather than solely a model-interpretation technique and provides a structured mechanism for integrating AI recommendations into Project Management Offices (PMOs), Stage-Gate governance, and project portfolio decision-making while preserving transparency, accountability, and meaningful human oversight.