Reducing Perishable Food Loss Through Forecast-Driven Stochastic Capacity Planning: Quantifying Value, Scalability, and Transfer Under Supply-Side Uncertainty

This work applies two-stage stochastic programming to perishable food loss driven by supply-side arrival uncertainty, a structure largely unaddressed in operations-research literature that focuses on demand-side uncertainty instead. Using real USDA shipment data across four commodities and multiple U.S. locations, we quantify the Value of the Stochastic Solution and Expected Value of Perfect Information with paired significance testing and cost-sensitivity analysis, find that a simple forecasting baseline is never significantly outperformed by gradient-boosted models, and show a genetic algorithm matches an exact solver within 2.83% while scaling substantially better at large scenario counts. A cross-location generalization test evaluates real-world transferability. The contribution is a rigorously validated, transparently reported decision framework rather than a claim of new theory — including where more sophisticated methods did not outperform simpler ones.