A Risk-Informed Framework for Sustainable Post-Conflict Urban Reconstruction: A Machine Learning Approach from Sana'a and Taiz.

This dissertation develops an innovative framework for rebuilding sustainably after conflict at the city level, using a combination of machine learning algorithms that utilize spatial data as well as multi-dimensional analytics of risk. It provides a mechanism for determining whether reconstruction has been successful through a predictive model based on random forests using 28 risk indicators across the domains of governance, socioeconomic, environment, and infrastructure. It also provides a unique district-level dataset spanning 2015–2026, containing data from remote sensing, government records, and field survey validation in 30 districts in Yemen. The prediction model produced an overall accuracy rate of 82% using cross-validation techniques that accounted for both spatial independence and data scarcity. Analysis of SHAP outputs suggests that governance fragility is the greatest risk-amplifying factor with the largest relative importance (i.e.,  0.34) and interacts with both environmental and socioeconomic stressors in a non-linear manner. The developed framework includes tools for quantifying uncertainty and simulating future scenarios, facilitating the dynamic prioritization of reconstruction interventions. By bridging predictive analysis and policy-oriented planning, this dissertation provides a framework for developing a decision support tool for equitably distributing resources to fragile, conflict-affected urban systems. Additionally,  it provides a novel methodology and practical guidance for building sustainable urban regions post-conflict in high-risk sites.