| Summary: | This dissertation investigates the advantages of optimization and machine learning algorithms to characterize, predict, and solve Response Planning and Preparedness problems in large-scale humanitarian organizations. Organizations often base their operational decisions on the staff's experiential knowledge rather than data-driven mechanisms. International professionals are one of the most valuable resources for humanitarian organizations that deliver food and relief items.
The problem of assigning such personnel to positions based on their preferences is a two-sided stable matching problem. Many two-sided markets form a matching between their agents by running centralized clearinghouse algorithms that ensure "stable" and "perfect" assignments. When dealing with large-scale organizations, agents often are not aware of all of their options or inquiring about all candidates/positions can be costly. It is a well-known fact that a perfect matching in a system
with partial information on agents' preferences is incompatible with ensuring stability. To address this issue, we design cycle-based approximation mechanism that models negotiations between self-interested agents and identifies a greater-cardinality matching by relaxing stability minimally while maximizing social welfare for all agents across multiple matching cycles. Similar to many other humanitarian organizations, UNHCR operates separate supply chains for the ongoing (OO) and
emergency relief (ER) operations, which is costly and susceptible to variances in ER demand quantities. We design network configuration capable of matching humanitarian relief supplies with uncertain demands for the UNHCR's supply chains. In doing so, a scenario-based two-stage stochastic modeling approach examines the possible actions for cost and lead time reduction such as merging OO and ER supply chains, moving global warehouses closer to demand locations, and incorporating
influencing factors in warehouse selection decisions. We employ a multi-criteria decision making approach that makes inventory pre-position, shipping, and transportation mode decisions. Shifting focus from reaction to anticipation of ER Response Planning and Preparedness, we propose a framework that improves the strategies for forecasting UNHCR's ER response as a zero-inflated skewed variable. Working within the proposed framework, we explore the relationship between countries'
characteristics and the UNHCR emergency relief response. We predict countries' future UNHCR ER response, the likelihood of different ER response levels, and the amount of ER response for a given country. In our analysis, identified patterns and associations determine the main drivers of ER response.
|
|---|
