UWSpace

UWSpace is the University of Waterloo’s institutional repository for the free, secure, and long-term home of research produced by faculty, students, and staff.

Depositing Theses/Dissertations or Research to UWSpace

Are you a Graduate Student depositing your thesis to UWSpace? See our Thesis Deposit Help and UWSpace Thesis FAQ pages to learn more.

Are you a Faculty or Staff member depositing research to UWSpace? See our Waterloo Research Deposit Help and Self-Archiving pages to learn more.

Photo by Waterloo staff

Communities in UWSpace

Select a community to browse its collections.

Recent Submissions

  • Thumbnail Image
    Item type: Item ,
    Evaluating Machine Learning Models for Predicting PFAS Adsorption onto Activated Carbon
    (University of Waterloo, 2026-09-23) Bui, Gia Thinh
    Per- and polyfluoroalkyl substances (PFAS) are a diverse class of persistent contaminants of increasing regulatory concern. Activated carbon (AC) adsorption is among the most widely implemented technologies for removing PFAS from water, with performance governed by the combined effects of PFAS physicochemical properties, AC characteristics, and water chemistry. Although isotherm experiments are commonly used to characterize PFAS adsorption on AC, they can be time-consuming and resource-intensive. Models capable of predicting PFAS adsorption could reduce the need for isotherm testing by guiding experimental design or replacing selected experiments. This thesis evaluated the applicability of machine learning (ML) regression models for predicting PFAS adsorption onto AC in two types of systems: (1) synthetic solutions containing a single PFAS and (2) real groundwater solutions containing multiple PFAS. For the single-PFAS system, isotherm data from 21 published studies were consolidated into a curated dataset comprising 658 observations across 11 PFAS and 36 AC materials. To prevent data leakage and rigorously assess model performance, an isotherm-aware data splitting approach was implemented alongside three complementary validation regimes, including 5-fold cross-validation (CV), leave-one-PFAS-out (LOPO) CV, and leave-one-adsorbent-out (LOAO) CV. Four features were selected to predict the solid-liquid distribution coefficient (log Kd): PFAS molecular weight, AC specific surface area, the difference between AC pHpzc and solution pH, and aqueous-phase PFAS concentration. Among the models evaluated, multiple linear regression demonstrated consistent predictive performance across all three validation regimes (5-fold CV: R² = 0.77, LOPO CV: R² = 0.76, LOAO R² = 0.75), comparable to more complex ML models. For the compiled dataset and selected feature space, these results indicate that the linear model offers a robust and interpretable alternative to more complex ML algorithms for predicting PFAS adsorption in relatively simple systems. The modelling approach was subsequently extended to a more complex problem of predicting multi-PFAS adsorption on colloidal activated carbon (CAC) in real groundwater. The dataset comprised 192 observations for PFOS, PFOA, PFHxS, PFHxA, and 6:2 FTS obtained from isotherm experiments conducted with seven groundwater samples collected from PFAS-impacted sites. In each experiment, the five PFAS were added to the groundwater at equimolar initial concentrations. Because the dataset was relatively small, the allocation of groundwater samples between training and test datasets was systematically varied to evaluate the sensitivity of model performance to data partitioning and differences in groundwater chemistry. Across the data splitting scenarios, extreme gradient boosting and random forest provided the most accurate and consistent predictions. However, both tree-based models showed systematic overprediction or underprediction in some data-split cases. Reserving adsorption data collected at a single CAC concentration for calibration corrected this systematic bias and substantially improved predictive performance. This calibration strategy provides a practical workflow in which a single-point experiment can be used to verify or adjust model predictions under new water chemistry conditions, thereby reducing the need for complete isotherm experiments. Overall, this thesis demonstrates the potential of data-driven models to support the characterization of PFAS adsorption in both single-PFAS synthetic solutions and multi-PFAS real groundwater solutions. The models can inform the design of bench-scale isotherm studies, prioritize experimental testing, and reduce the experimental effort required to evaluate AC-based PFAS treatment. The contrasting model choices across the two studies further demonstrate that the predictive advantage of complex ML models over linear regression is situational rather than universal and should be evaluated rather than assumed. Beyond providing practical predictive tools, this work illustrates rigorous approaches to data splitting and model evaluation for relatively small experimental datasets, providing a framework for strengthening the reliability of data-driven modelling of contaminant adsorption.
  • Thumbnail Image
    Item type: Item ,
    Gradient-based Methods for Multi-Objective Optimization with Applications in Machine Learning
    (University of Waterloo, 2026-09-23) Hu, Zeou
    Many machine learning problems involve trade-offs among multiple objectives, such as accuracy, fairness, or the interests of different tasks or users, making multi-objective optimization (MOO) a natural framework for their study. Such trade-offs arise in a range of modern machine learning settings, including but not limited to multi-task learning, federated learning, algorithmic fairness, and reinforcement learning. While MOO has long been studied in the optimization literature, often through classical approaches such as evolutionary algorithms, contemporary machine learning problems are typically high-dimensional and call for scalable gradient-based methods. This thesis studies gradient-based MOO from three complementary perspectives: its application to federated learning as an important machine learning setting, the refinement of its solution concepts under variable sparsity, and the development of a unifying theory for gradient aggregation methods.
  • Thumbnail Image
    Item type: Item ,
    Catastrophe Risk Management in Climate Change
    (University of Waterloo, 2026-09-23) Guo, Yuxuan
    Climate change challenges several assumptions underlying the insurability of catastrophe risks. Losses can be heavy-tailed, multiple participants can be exposed to the same systemic climate drivers, and the severity of extreme losses must be assessed from limited and potentially heterogeneous data. These features weaken conventional diversification and make both the design and quantitative assessment of insurance mechanisms more difficult. This thesis develops an integrated framework for managing catastrophe risk under these challenges in three connected steps: (i) diagnosing when risk pooling succeeds or fails, (ii) designing pooling mechanisms when simple pooling is insufficient, and (iii) improving the reliability of these decisions by accounting for uncertainty in the estimation of tail risk. In Chapter 3, we study the limits of diversification under a general additive heavy-tailed factor model in which each participant’s loss consists of a common systemic component with heterogeneous and potentially random exposure and an idiosyncratic component. We derive participant-level tail and diversification ratios that compare a participant’s risk capital requirement when participating in the pool with the corresponding requirement when standing alone, and characterize conditions for pool viability under different relative tail regimes for the systemic and idiosyncratic factors. The results show that when systemic risk dominates, diversification depends critically on the dependence structure of participants’ exposures. In particular, complementary exposure structures generate greater diversification, while in growing pools the limiting diversification benefit depends jointly on the relative growth of the loss threshold and pool size, and on the concentration of participants’ systemic exposures. Motivated by the possibility that pooling alone may fail to provide a diversification benefit under heavy-tailed systemic risk, Chapter 4 treats risk pooling as a design problem involving three complementary mechanisms: assessment design, participant selection, and external tail-risk transfer. For a fixed pool, we derive minimax assessment rules that maximize the diversification benefit of the least advantaged participant and characterize pool viability. Then we study optimal pool composition, introducing a diversification index to guide participant selection according to the complementarity of their systemic exposure profiles. Finally, we examine the effect of an insurance-linked securities layer on finite-level diversification. Simulation results demonstrate that combining these mechanisms improves pool viability exposed to heavy-tailed systemic risk. The implementation of these pooling and design decisions depends critically on the reliable estimation of tail heaviness. However, conventional tail-index estimators can be sensitive to contamination, heterogeneity, threshold selection, and slow convergence to the asymptotic tail model. Therefore, Chapter 5 develops Wasserstein distributionally robust versions of the Hill estimator and a power-based tail-index estimator. The proposed estimators explicitly account for distributional uncertainty around the empirical distribution. Simulation studies show that the distributionally robust estimators reduce downside underestimation under contaminated-mixture distributions and distributions with slow convergence to their asymptotic tails, although this protection may be accompanied by additional conservatism under simple Pareto cases. An application to Norwegian fire insurance claims illustrates how the proposed distributionally robust estimated tail indices and extreme quantiles could be used as conservative, stress-adjusted assessments of extreme-loss risk. In addition to the quantitative study of physical climate risk in the main chapters, this thesis considers climate liability risk as a complementary dimension of climate risk management. Appendix A examines the measurement and modeling of climate liability risk arising from climate-related litigation, regulatory and disclosure obligations, and evolving standards of corporate responsibility associated with the transition to a low-carbon emission. This complementary study broadens the perspective of the thesis from the management of physical catastrophe losses to the wider range of risks that climate change creates for insurers and other organizations. The thesis shows that managing heavy-tailed systemic risk requires more than increasing the scale of risk pooling. Effective risk management requires identifying when systemic risk is diversifiable, redesigning pooling arrangements in response to the limits of diversification, and ensuring that these decisions remain reliable under uncertainty about the underlying tail behavior. The thesis thereby connects the diagnosis, design, and estimation problems within a unified framework for catastrophe risk management under climate change.
  • Thumbnail Image
    Item type: Item ,
    Integrating Sustainability Assessment into Project-Level Urban Development Planning to Support Regenerative Urban Outcomes
    (University of Waterloo, 2026-09-23) Mohamed, Hanan
    This thesis examines how project-level urban development planning and alternatives analysis can more systematically integrate Sustainability Assessment to support regenerative urban outcomes. It develops and applies a 15-dimensional Context-Adaptive Sustainability Assessment Framework grounded in Gibson’s sustainability requirements and informed by systems thinking, regenerative sustainability, post-normal science, pluriversal governance, antifragility, and sustainable project management. The study uses an exploratory qualitative multiple-case study design focused on five urban development projects in Waterloo Region, Ontario: evolv1, Ahrens Street West, One Hundred Victoria Street South, Northfield Station Area Development, and Breithaupt Block. The framework is applied retrospectively through document analysis, narrative-based assessment, semi-quantitative ordinal rating, integration mapping, and researcher-constructed Document-Derived Causal Mapping with Feedback Loop Analysis. The reconstructed maps represent directed, positive or negative influence relationships among factors and are developed from coded literature and project documents rather than direct participant input; they are therefore used as interpretive models rather than as validated proof of causality. This approach enables structured comparison across diverse project types while remaining responsive to governance conditions, planning maturity, stakeholder engagement, and the availability of evidence. Findings show that sustainability integration across the five projects was uneven and dimension-specific. Environmental, technological, spatial, and selected economic dimensions were generally more visible, particularly where technical standards, policy alignment, or measurable performance criteria supported implementation. In contrast, equity, affordability, cultural inclusion, ethical deliberation, life-cycle learning, and adaptive governance remained consistently underdeveloped. The reconstructed document-derived causal maps further suggested that the documented project logics were predominantly linear, with no closed feedback pathways identified in the five project-specific maps and limited documentary evidence of mechanisms for ongoing learning. The thesis offers a flexible, governance-aware, and learning-oriented framework intended to strengthen the early integration of sustainability considerations into urban development planning. It reframes Sustainability Assessment as a diagnostic and decision-support process that begins at project conception rather than functioning as a late-stage compliance mechanism.
  • Thumbnail Image
    Item type: Item ,
    Understanding the Complex Nexus between Government Stringencies and Adverse Public Sentiments and Misinformation in Social Media: Effects on Pandemic Preparedness and Response
    (University of Waterloo, 2026-09-23) Abdullah, Abu Yousha Mohammad
    The COVID-19 pandemic highlighted the need for surveillance approaches that account for both epidemiological trends and the online information environment. Government policies, misinformation, and public sentiment evolved alongside pandemic conditions, potentially providing complementary information for forecasting. This study examined whether lagged government policy indicators and social-media signals improved short-term forecasts of COVID-19 cases and deaths in Canada beyond recent epidemiological history. An ecological time-series study integrated national COVID-19 outcomes, twelve Oxford COVID-19 Government Response Tracker policy indicators, and a corpus of 5,971 Reddit posts from January 2020 through December 2021. A stratified sample was manually annotated for misinformation and negative sentiment. DeBERTa-v3 and RoBERTa classifiers were fine-tuned to generate post-level misinformation and sentiment predictions, while a BART-based zero-shot classifier characterized discussion topics. Social-media indicators were aggregated at weekly and biweekly resolutions. Separate XGBoost models predicted cases and deaths using four predictor configurations: epidemiological indicators alone, epidemiological and policy indicators, epidemiological and social-media indicators, and all predictor groups combined. Models were evaluated using chronological training, validation, and test partitions. SHapley Additive exPlanations (SHAP) characterized feature contributions. The misinformation classifier demonstrated high precision but limited sensitivity, while the sentiment classifier showed stronger ranking performance than its threshold-based classification results. Vaccine discussions and lockdowns and restrictions dominated the selected topic-analysis subset. Forecasting performance varied by outcome and temporal resolution. Weekly mortality forecasting showed the clearest improvements from additional policy and social-media indicators. For biweekly mortality, the epidemiological-plus-policy model performed best, and adding social-media indicators did not improve performance further. Case forecasting remained predominantly dependent on recent epidemiological history, with inconsistent improvements for weekly cases and poorer performance from augmented models for biweekly cases. Although social-media features ranked highly in some SHAP analyses, their importance within fitted models did not consistently translate into improved test-set accuracy. These findings suggest that policy and Reddit-derived social-media indicators may complement conventional epidemiological surveillance, particularly for mortality forecasting. However, the small temporal samples, selective Reddit coverage, and annotation and classification uncertainty limit generalizability. The results represent retrospective predictive associations rather than causal effects or demonstrated operational early-warning capability. Further evaluation across forecast periods, geographic settings, and more representative data is needed to establish their practical value for pandemic preparedness and response.