Methods of Training a Gamma Mixture Hurdle Model for Estimating Corresponding Food Flows between Regions

An article of manufacture including a non-transitory computer-readable medium, having stored thereon program instructions that, upon execution by a computing system, cause the computing system to perform operations comprising:

• training a gamma mixture hurdle model characterizing a functional relationship between:
• output data specifying food flows between zones, and input variables representing food production and food consumption in the zones, wherein the training involves:
• (i) using binary logistic regression to estimate whether corresponding food flows exist between pairs of the zones, and (ii) for pairs of the zones in which the corresponding food flows are estimated to exist, using a gamma mixture model to estimate amounts of the corresponding food flows, and wherein each zone includes a respective set of regions;
• based on parameters of the gamma mixture hurdle model, estimating:
• (i) whether corresponding food sub-flows exist between pairs of the regions, and (ii) for pairs of the regions in which the corresponding food sub-flows are estimated to exist, potentials of the corresponding food sub-flows;
• and
• determining, using a linear programming framework, values for the corresponding food sub-flows, wherein the linear programming framework uses data specifying food flows as mass balance constraints on totals of the corresponding food sub-flows that are within each of the zones, the potentials as inequality constraints on the corresponding food sub-flows that are within each of the zones, and region-to-region distance data for the regions as a minimization constraint on the corresponding food sub-flows.

A computing system comprising:

• one or more processors;
• memory;
• and
• program instructions, stored in the memory, that upon execution by the one or more processors cause the computing system to perform operations comprising:
• training a gamma mixture hurdle model characterizing a functional relationship between:
• output data specifying food flows between zones, and input variables representing food production and food consumption in the zones, wherein the training involves:
• (i) using binary logistic regression to estimate whether corresponding food flows exist between pairs of the zones, and (ii) for pairs of the zones in which the corresponding food flows are estimated to exist, using a gamma mixture model to estimate amounts of the corresponding food flows, and wherein each zone includes a respective set of regions;
• based on parameters of the gamma mixture hurdle model, estimating:
• (i) whether corresponding food sub-flows exist between pairs of the regions, and (ii) for pairs of the regions in which the corresponding food sub-flows are estimated to exist, potentials of the corresponding food sub-flows;
• and
• determining, using a linear programming framework, values for the corresponding food sub-flows, wherein the linear programming framework uses data specifying food flows as mass balance constraints on totals of the corresponding food sub-flows that are within each of the zones, the potentials as inequality constraints on the corresponding food sub-flows that are within each of the zones, and region-to-region distance data for the regions as a minimization constraint on the corresponding food sub-flows.