Improving Food and Logistics Systems Through Analytics
Name
zhang-eyzhang-phd-orc-2026-thesis.pdf
Size
4.1 MB
Format
Adobe PDF
Checksum (MD5)
7ec2438c3b4968df131aea7dbb5104ce
Author(s)
Zhang, Emily
Advisor(s)
Levi, Retsef
Perakis, Georgia
Date Issued
May 2026
Publisher
Massachusetts Institute of Technology
Abstract
Food systems are marked by significant inefficiencies, with substantial food waste coexisting alongside food insecurity. While various public health policies and industry practices aim to address these intertwined issues, they are often not rigorously evaluated or optimized to mitigate sources of uncertainty and address heterogeneous conditions. This thesis develops new models and algorithms to improve operational decisions of companies and evaluation of public health policies and programs in food systems, with a particular focus on inventory planning and causal inference methodologies.
Chapter 2 studies the issue of food waste in retail settings, where inventory decisions must be made under demand uncertainty. The chapter describes and analyzes a reservation-based system that allows consumers to reserve products in advance at a discounted price, thereby revealing partial information about future demand. The proposed modeling approach captures consumer shopping behavior and the analysis shows that, under realistic conditions, this system can simultaneously reduce waste and increase retailer profit. The analysis further characterizes the settings and product types for which these benefits are most pronounced, providing guidance for implementation.
Chapter 3 considers the complementary challenge of insufficient food access through a partnership with the Massachusetts Healthy Incentives Program (HIP), a state-level program that complements the federal SNAP program by offering rebates on purchases of fruits and vegetables at participating vendors. The chapter examines how data can be leveraged to better evaluate and design public interventions, particularly in settings where impacts are heterogeneous across consumers and geographic locations. In particular, the chapter is concerned with how the expansion of vendor networks affects program utilization and how these effects vary across populations. To address the methodological challenges posed by irregular intervention timing and heterogeneous impacts, the chapter develops the Panel Clustering Estimator (PaCE), a novel approach for treatment effect estimation in panel data. The analysis establishes theoretical guarantees for the estimator and demonstrates improved performance relative to existing methods. Applying PaCE to large-scale transaction data uncovers substantial heterogeneity in the impacts of vendor additions across the state and provides insights for future vendor expansions.
Finally, Chapter 4 considers a multi-item inventory optimization problem with joint ordering costs. Specifically, the chapter studies the joint replenishment problem (JRP) under a broad class of submodular ordering cost functions that are defined based on a decomposition of the items into k categories, where the cost depends on weighted aggregate quantities within each category while allowing for general interactions across categories through a joint cost function. The main result in this chapter is an approximation algorithm that achieves an (8k+1)-approximation guarantee, providing the first constant-factor guarantee for this broad class of cost functions. The algorithm and the worst-case performance analysis are based on a novel water-filling procedure for rounding a linear programming relaxation, and the resulting structural insights may inform future work on inventory problems with submodular cost structures.
MIT Department
Massachusetts Institute of Technology. Operations Research Center
Sloan School of Management
Terms of Use
Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0)
Copyright retained by author(s)
Persistent DSpace Link
