Optimization under uncertainty for food security (doctoral thesis)

Thirty percent of the global population experiences food insecurity due to a lack of sufficient, affordable, and nutritious food, preventing them from living healthy and active lives. Through mathematical optimization and collaboration with food assistance programs, this thesis provides possible solutions to address the complexities and uncertainties of real-world challenges in food security. Methods of optimization under uncertainty, including robust optimization, stochastic optimization, inverse optimization, and tree-based machine learning, are explored and applied to problems arising in three specific food assistance programs. The first two programs are food bank organizations: the Association of Dutch Food Banks (the Netherlands) and Moisson Montréal (Canada), for which optimization methods for investment and routing challenges are studied. For the third program, the United Nations World Food Programme, applications of machine learning provided estimates of the number of children under five suffering from acute malnutrition. In addition to solving real problems faced by these food assistance programs, this thesis advances theory in optimization under uncertainty. A matheuristic is presented that finds feasible solutions for the vehicle routing problem when demand, service, and waiting times are stochastic. Furthermore, a convex reformulation for a class of nonconvex optimization problems is introduced, providing results that are useful in many fields, including inverse optimization and robust optimization.