Theses and Dissertations

ORCID

0009-0000-8975-5992

Advisor

Lee, Seunghan

Committee Member

Ma, Junfeng

Committee Member

Smith, Brian

Committee Member

Wang, Haifeng

Date of Degree

8-12-2026

Original embargo terms

Embargo 2 years

Document Type

Dissertation - Open Access

Major

Industrial & Systems Engineering

Degree Name

Doctor of Philosophy (Ph.D.)

College

James Worth Bagley College of Engineering

Department

Department of Industrial and Systems Engineering

Abstract

Natural disasters, including hurricanes and wildfires, necessitate comprehensive evacuation planning and long-term mitigation strategies, which rely heavily on an understanding of decision-making processes under uncertainty. The disaster management cycle consists of four phases: mitigation, preparedness, response, and recovery. Decisions made in one phase significantly influence vulnerabilities and capacities in subsequent phases. Therefore, effective disaster management requires models that consistently represent human behavior across all four stages, rather than treating each phase in isolation. Current simulation methods typically depend on a singular modeling approach and static behavioral assumptions, which restrict their capacity to simulate how decision-making evolves with changing conditions. This dissertation seeks to fill that gap by developing a data-driven hybrid simulation framework that directly integrates behavioral and sensor data into structured modeling architectures, enabling models to dynamically represent human decision-making throughout all phases of disaster management. First, it formalizes a process for converting unstructured behavioral signals, such as sentiments expressed in social media posts and post-event survey responses, into decision rules within an agent-based model utilizing emotion classification and basket analysis. This allows for co-occurring emotional states to influence individual agent behavior, rather than relying on fixed behavioral assumptions. To incorporate this into human decision-making, Ornstein-Uhlenbeck processes are employed to simulate evacuation choices over time in an agent-based model under the Overview-Design-Details modeling framework. Second, the dissertation introduces a hybrid architecture that combines system dynamics with agent-based modeling, enabling population-level trends and heterogeneous individual-level factors to inform and interact with each other at each timestep. This approach extends hybrid simulation methods to the long-term implementation of governmental policies, which are crucial for disaster mitigation support. Third, it applies a hidden Markov model to infer latent system states, such as traffic congestion, from indirect sensor data, illustrating how hidden-state estimation can utilize information when disrupted by conditions caused by natural disasters. Collectively, this dissertation establishes a generalizable methodology for embedding behavioral and sensor data into simulation frameworks, providing a transferable approach for modeling human decision-making in socio-technical systems beyond these specific hazards and into other time-sensitive, high-duress disaster scenarios.

Download

Available for download on Saturday, September 09, 2028

Share

COinS