This study presents a hybrid predictive framework for optimizing retail supply chains under demand and quality uncertainty. Data from a multi-store retailer in the Marmara Region of Türkiye is used. In the predictive phases, machine learning models such as XGBoost (XGB), Gradient Boosting (GB), Decision Tree (DT), and Linear Regression (LR) are utilized to generate demand scenarios and estimate defect risk with limited data. To ensure the robustness and processability of the model presented in the framework, the predictive outputs are converted into deterministic parameters. The mixed integer linear programming (MILP) model, one of the phases in the framework, aims to minimize logistics and operational costs while addressing capacity, flow balance, and service level constraints. Scenario analysis shows that considering demand uncertainty and defect risk significantly impacts shipping decisions and costs. The proposed framework offers a flexible and scalable skeleton for risk-conscious retail logistics planning.

Keywords

Predictive–prescriptive analytics, Retail supply chain management, Mixed-integer linear programming (MILP), Demand and quality uncertainty, Machine learning in operations research