Date of Award

8-2026

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Industrial Engineering

Committee Chair/Advisor

Amin Khademi

Committee Member

Qi Luo

Committee Member

Thomas Sharkey

Committee Member

Yongjia Song

Abstract

This dissertation develops theory and algorithms for sequential decision-making under uncertainty, with applications to disaster response and patient-centric clinical trials. The first study addresses evacuation shelter opening during hurricanes. It formulates shelter activation as a multi-class optimal stopping problem with irreversible, priority-constrained decisions and proposes a direct policy approximation method based on hierarchical neural networks. A Hurricane Florence case study shows that the proposed policy can reduce response costs by adapting shelter opening times to evolving, physics-informed forecasts. The second study examines when a fully informed patient should enroll in an early-stage clinical trial. Using a Bayesian optimal stopping model, it captures how beliefs about treatment toxicity and efficacy evolve while patient health deteriorates, and establishes control-limit enrollment policies that balance immediate treatment benefit against the value of waiting. The third study extends the enrollment-timing problem to a multi-patient setting in which prospective participants observe shared interim trial outcomes. It models enrollment as a dynamic information-sharing game, characterizes Markov perfect equilibrium behavior, and develops computational methods for larger instances. The analysis shows that transparency can improve patient welfare through learning, but may also induce strategic waiting. Together, these studies provide structured, interpretable, and computationally viable decision policies for high-stakes systems where uncertainty, irreversibility, learning, and strategic interaction shape real-time decisions.

Author ORCID Identifier

C16863574

Download

Available for download on Tuesday, August 31, 2027

Share

COinS