Artificial intelligence (AI) is increasingly transforming recruitment and talent-acquisition processes; however, concerns regarding opaque predictions, unreliable confidence estimates, algorithmic bias, and inadequate human oversight continue to limit its responsible adoption. This study proposes an explainable and confidence-adaptive AI framework for identifying potentially problematic recruitment patterns in online job postings while retaining human control over uncertain decisions. A publicly available corpus of 22,000 U.S. job advertisements was analyzed, of which 21,879 records remained after quality screening. Weak-supervision rules were developed around four recruitment-risk dimensions: role alignment, experience rationality, workload feasibility, and internal discrepancy. Five machine-learning models—Logistic Regression, Support Vector Machine, Random Forest, XGBoost, and LightGBM—were compared using a leakage-controlled evaluation protocol. LightGBM achieved the strongest validation performance and was subsequently calibrated using sigmoid calibration with a validation-optimized decision threshold. On the held-out test set, the final model achieved 87.08% accuracy, 87.25% precision, 97.16% recall, 91.94% F1-score, 0.621 Matthews correlation coefficient, 0.939 ROC-AUC, and 0.980 PR-AUC. A confidence-adaptive Human-in-the-Loop mechanism routed 44.98% of cases for review while capturing 96.30% of observed model errors. SHAP explanations, multidimensional risk profiling, semantic-hybrid representation, and potential bias-trigger screening further improved interpretability and auditability. The proposed system is positioned as recruitment-content decision support rather than an autonomous hiring mechanism.

Tilottama Azad,"Intelligent Screening of Online Job Postings for Recruitment Risk and Responsible Hiring Support" International Journal of Advanced Multidisciplinary Research and Educational Development, V2(5): Page(413-419) September - October 2026. ISSN: 3107-6513. www.ijamred.com. Published by Scientific and Academic Research Publishing.