Predictive–Prescriptive Workforce Scheduling Under Absenteeism Uncertainty: A Computational Proof of Concept with Saudi-Relevant Logistics Scenarios

Absenteeism-related workforce uncertainty complicates logistics scheduling. This study evaluates a leakage-aware predictive–prescriptive proof of concept, using 740 observations from 36 workers in the Brazilian UCI Absenteeism at Work dataset. The measured outcome is whether recorded absence duration exceeds two hours, not future shift nonattendance. Logistic Regression, Random Forest and XGBoost were assessed with worker-disjoint testing, grouped tuning, calibration and cluster bootstrap uncertainty. Holdout ROC-AUC values were 0.4959, 0.4765 and 0.5243, respectively, indicating weak discrimination. A scheduling layer evaluated hypothetical availability-risk scenarios, using CP-SAT and mixed-integer programming with shortage, overtime, rest and fairness constraints. The solvers agreed on a small verification instance, and CP-SAT returned feasible schedules for 100 workers over 14 days and 200 workers over 30 days, with 200-worker optimality gaps near 19–21%. Across five replicated scenarios, shortage priority raised modelled coverage over greedy scheduling by 0.01408, whereas overtime priority reduced it by 0.07901. A value-of-prediction test on identical rosters compared forecast, uninformative, permuted and oracle risk inputs: the Logistic Regression forecast raised the modelled coverage by 0.0115 over an uninformative input (about 20% of the oracle gain), but this was not distinguishable from zero under worker-level resampling; the XGBoost forecasts lowered coverage, and changing objective weights moved coverage approximately eight times as much. The Saudi-labelled scenarios are hypothetical, not Saudi observations. The study is a methodological, limited-value assessment, not evidence of Saudi supply-chain resilience.