Skip to content
Publications

Article · International Journal of Industrial Engineering and Management · September 1, 2026

Learning Equitable and Efficient Shift Rosters with Multi-Objective Deep Reinforcement Learning and Counterfactual Fairness Constraints in Operations

Bùi Thành Khoa, Kodirbek Makharov, Samandarov Batirbek Satimovich, Sanatbek Jafarov, Dilshod Raupov, Otabek Djurabaev

Abstract

Shift rostering systems often optimize cost and coverage while treating fairness as a secondary objective. This study develops a multi-objective deep reinforcement learning framework for workforce scheduling that integrates Proximal Policy Optimization with a graph neural network encoder, a Conditional Value-at-Risk term for limiting extreme burden concentration, and a counterfactual fairness regularizer for reducing demographic sensitivity in assignment decisions. The framework was evaluated on 18 months of workforce data from a tertiary hospital and an industrial facility in Uzbekistan. Relative to mixed-integer linear programming and vanilla reinforcement learning baselines, the proposed model reduced the burden Gini coefficient by 41.3%, narrowed weekend disparity by 36.5%, and achieved adverse impact ratios above 0.97 for gender and age. These equity gains were obtained with a 1.8% increase in operational cost and a 9.4% reduction in understaffing minutes under demand shocks. The results indicate that fairness constraints can be embedded directly into policy learning while preserving service levels and improving resilience in high-constraint scheduling environments.

Scheduling and Timetabling SolutionsHealthcare Operations and Scheduling OptimizationScheduling and Optimization Algorithmsshift schedulingworkforce schedulingcounterfactual fairness

More by Batirbek Samandarov