Abstract
Retailers face significant challenges in workforce scheduling due to highly variable demand, rising labor costs and strict service requirements. This study presents a novel, two-stage optimisation framework that integrates ensemblebased machine learning forecasting with the mathematical programming approach CPLEX for shift scheduling in the retail sector. In the first stage, advanced machine learning models (XGBoost, Random Forest and SARIMAX) generate probabilistic foot traffic forecasts by leveraging extensive feature engineering. These forecasts are then used as inputs for a shift optimization model that considers complex operational constraints, such as labor laws, union agreements and employee preferences. Deployment in a major Turkish retail chain yielded empirical results demonstrating an 11.3% reduction in labor costs, a 47% decrease in understaffing, while maintaining full coverage with minimal excess staffing, and an 8.2% improvement in service levels across more than 200 stores. The proposed approach yields scalable cost savings and service enhancements, providing a robust decision-support tool for workforce management in dynamic retail environments.
Showing the abstract — retrieve the full paper via the Exa API.