Data Science Commons^™

The Influence Of Scale In Modeling Social Vulnerability And Disaster Assistance, Sina Razzaghi Asl, Oronde Drakes, Eric Tate, Samuel Brody, Wesley Highfield, Kayode Atoba

Political Science & Geography Faculty Publications

Understanding how social vulnerability relates to disaster impacts is critical for addressing social equity, yet the role of spatial scale in this relationship is often overlooked. Most studies use aggregated data, risking ecological fallacy-misinterpreting individual outcomes from group-level data. This study examines how spatial scale influences the relationship between social vulnerability and federal disaster assistance after Hurricane Harvey. Using spatial econometric models at both household and census tract levels, we assessed the strength of key vulnerability indicators in explaining disaster assistance. Results show that disability, housing tenure, household size, and income predict assistance at the household level, but their influence …

Blens: Biomedical Literature Extraction And Scoring System, Tyler J. Simone

Honors Theses and Capstones

Systematic reviews and meta-analyses represent the gold standard for evidence synthesis in healthcare, yet their manual execution remains labor-intensive, time-consuming, and vulnerable to human bias. With the exponential growth of biomedical literature, traditional literature screening and analysis has become increasingly unstable and noncomprehensive. This thesis presents the development and validation of an automate literature gathering and review system that integrates multiple scientific databases through a unified desktop application. The platform combines APIs from PubMed (NCBI Entrez), ClinicalTrials.gov, bioRxiv and medRxiv to enable simultaneous, standardized searching across peerreviewed and preprint sources. Built in Python with a PySide6 graphical interface, this standalone …

Previsit Ai: A Retrieval-Augmented Generation For Patient Readiness In Clinical Encounters, Rolande Umuhoza

All Graduate Theses, Dissertations, and Other Capstone Projects

With healthcare systems under growing pressure from rising patient volumes and shrinking consultation windows, improving how patients communicate with physicians has become essential to delivering quality care. Yet patients routinely arrive at appointments unable to clearly describe their symptoms, recall their medical history, or articulate concerns, contributing to miscommunication, diagnostic inefficiency, and pre-visit anxiety. This study introduces PreVisit AI, a conversational system designed to address this gap through structured, knowledge-based patient preparation. The system is built on a Retrieval-Augmented Generation (RAG) architecture combining HuggingFace sentence embeddings (all-MiniLM-L6-v2), a Chroma vector store, and Google’s Gemini language model over a curated seven-document …

Impact Of Cross-Client Heterogeneity In Federated Learning For Real-World Plant Disease Classification, Ritesh Janga

All Graduate Theses, Dissertations, and Other Capstone Projects

Deep learning applications are being adopted in agricultural image analysis that include challenges of data privacy and limited institutional data and heterogeneity of different types of architectures. However, Federated Learning is a model that allows collaborative training on data that does not have to be shared among parties. Therefore, Federated Learning is an effective method of collaborative training; however, its comparative effectiveness as compared to individual (local) training on diverse architectures has never been examined in an agricultural context. The objective of this study was to examine Federated Learning for the purpose of crop disease classification on extreme non-IID distributed …

Toward Interpretable Multi-Omics Multimodal Biomedical Artificial Intelligence, Yanjun Lyu

Computer Science and Engineering Dissertations

The complexity of human disease arises from biological processes that unfold across multiple scales, from molecular variation through cellular function, tissue organisation, brain phenotypes, each of which is associated with distinct measurement modalities, regularities, and characteristic. Contemporary biomedical artificial intelligence has brought the opportunity to reveal the complexity with in; however, its methodological default, in which models are trained on most readily available modality, does not adequately engage with the multi-scale connected structure by which biological meaning is constituted. The research area of multi-omics and multi-modal AI for biomedicine remains at an early exploratory stage, and the work presented in …

Predicting Criminal Behavior In Major Us Cities, Madison A. Price

SPARK Symposium Presentations

In recent years, especially post pandemic, there has been a decrease in crime in the United States. Unfortunately, the country’s violent crime rates are still significantly higher compared to similar high-income countries, so what predicts crime in major American cities? There is tons of research to support the idea that demographics can offer some insight into predicting crime. There are countless online resources that seek to identify major crime centrals in the United States (Petrino, 2025). In the late 1990s, researchers noticed that crime rates in cities had a downward slope due to an important contributor: demographic change (Fox & …

Validity Assessment Of Resting Heart Rate Variability From The Garmin Health Snapshot, Kayla M. Porter, Andrew Flatt

Honors College Theses

Purpose: To assess the agreement between the Garmin Forerunner 265 (a commercially available sports watch) and a single-channel electrocardiographic (ECG) chest strap for determining resting heart rate variability (HRV). Secondary aims were to assess the impact of skin tone and body position on measurement accuracy.

Methods: Young adults (n = 30, 57% women) aged 18–39 years without known cardiovascular conditions and without tattooing or scarring on the dorsal left wrist were recruited. HRV was recorded simultaneously using ECG and the Forerunner 265’s optical sensor during Garmin’s 2-minute “Health Snapshot.” Measurements were obtained in three standardized positions: supine, seated, and standing. …

Data-Driven Climate Damage Functions For Capital Formation: Estimating The Climate Penalty Using Maching Learning, Pramudya Wicaksono

All Graduate Theses, Dissertations, and Other Capstone Projects

Traditional integrated assessment models assume parametric climate damage functions that may miss nonlinearities, heterogeneity, and dynamic effects on investment. This thesis develops a data-driven climate damage function for capital formation by estimating the predictive relationship between climate conditions and future gross fixed capital formation (% GDP) across 125 countries over 1982–2019. We construct a panel dataset by combining daily ERA5 climate reanalysis data (accessed via the Copernicus Climate Data Store API and aggregated to yearly country-level variables including temperature anomalies, extreme heat days, frost days, precipitation, and solar radiation) with economic indicators from the World Bank World Development Indicators and …

Data Defines Success: Algorithm For Dataset Quality Assessment In Deep Learning For Malware Detection, Matei Ionescu

Theses and Dissertations

The field of artificial intelligence is based upon the premise of constructing architectures through which to propagate training data. However, the majority of existing research literature is focused on architecture. While necessary, the attention devoted to the architecture should not so precipitously exceed that of the data. It should be noted that this disparity is not without reasonable cause. Data quality is often exceedingly difficult to verify due to particularities of the field or subfield; LLM repositories of text are distinct from image recognition pictures of dog breeds which are distinct from EEG waveforms of human brains which are distinct …

Neighborhood Embeddings And Scalable Learning For Optimal Transport And Unbalanced Optimal Transport, Muhammad S. Rana

Mathematics Dissertations - Archive

Dimensionality reduction techniques are developed from the assumption that high-dimensional data often arises from low-dimensional structures embedded into the high-dimensional ambient space. Classical dimensionality reduction methods rely on the Euclidean distance, which may fail to capture the geometric structures of the datasets. This dissertation includes alternative metrics for dimensionality reduction techniques and challenges in applying these techniques.

First, we investigate the Wasserstein distance based neighbor embeddings for dimensionality reduction methods and compare the classification and clustering performance with the classical Euclidean based methods. The Wasserstein distance models the data as probability distributions and compares two distributions applying optimal transport (OT) …

Solving High-Dimensional Differential Equations Using Recurrent And Residual Neural Network Architectures, Hind Khaled Kolaib

Knowledge Engineering and Data Science

High-dimensional Partial Differential Equations (PDEs) form the foundation of complex process modeling in various scientific and engineering applications, including finance, physics, and optimal control. However, classical numerical methods are adversely affected by the curse of dimensionality, making them inapplicable for large-scale problems. Recently, however, deep learning-based approaches have provided a new toolbox for these high-dimensional PDEs, including methods such as the Deep Backward Stochastic Differential Equation (Deep BSDE) method. Our approach draws on a more sophisticated deep learning backbone, using neural networks (in our case, a Residual Neural Network and a Long Short-Term Memory network (LSTM) integrated into the Deep …

Analysis Of Δ¹¹B As A Seawater Ph Proxy: Comparing Ocean Circulation Inverse Model Output With Marine Calcifier Geochemistry, Jesse I. Dong

CMC Senior Theses

Increasing anthropogenic carbon flux into the oceans decreases seawater pH, alters dissolved inorganic carbon speciation, and reduces biogenic calcification. The marine calcifiers— specifically corals and coralline algae—incorporate elements from surrounding seawater into their carbonate structures, which preserve past records of ocean carbon chemistry. In particular, boron in biogenic carbonate is a potentially valuable proxy for historical ocean pH across human timescales. Within seawater, boron primarily exists as boric acid B(OH)3 and borate ions B(OH)4 - , where higher pH favors the formation of borate ions. Borate ions preferentially incorporate the heavier ¹¹B isotope over 10B. On the other hand, if …

Adversarial Robustness In Biomedical Time-Series Models, Rohan Tiwari

Bioengineering Theses

This study investigates adversarial vulnerabilities in deep learning models for biomedical time-series classification across two clinically important modalities: electrocardiography (ECG) and electroencephalography (EEG). Using the MIT-BIH Arrhythmia and CHB-MIT seizure datasets, I evaluate time-domain attacks (FGSM, PGD), Fourier-domain constrained attacks, and learned spectral perturbations designed to reveal modality-specific sensitivity patterns. Across both tasks, a consistent trend emerges low-frequency components (0–5 Hz) constitute a dominant axis of adversarial vulnerability, with perturbations in this range producing the steepest degradation in classification performance. In ECG models, protecting the physiologically relevant QRS band (5–20 Hz) significantly improves robustness, whereas EEG models remain highly sensitive …

Microbial Community Structure In Global Soils, Matthew Jabro

CMC Senior Theses

Soil harbors the most diverse microbial communities on Earth, yet whether predictable community types exist across biomes and whether taxonomic composition encodes habitat of origin remain open questions at global scale. This thesis addresses both questions by applying unsupervised clustering and supervised classification to transformed 16S ribosomal RNA (rRNA) amplicon profiles from two independent datasets: the global topsoil survey of Bahram et al. (193 samples) and the Earth Microbiome Project (EMP) soil subset of Thompson et al. (2,209 samples). Application of a sample clustering method based on a mixture of Gaussian Graphical Models (MixGGM) identified 19 clusters in the topsoil …

Network-Aware Airline-Specific Flight Delay Prediction Using Tree-Based Ensemble Models, Mary Dufie Afrane

College of Graduate Studies: Theses & Dissertations

Flight delays pose persistent challenges to the efficiency and reliability of air transportation systems, affecting airlines, airports, regulators, and passengers alike. As traffic demand grows and operational environments become increasingly interconnected, accurately predicting both departure and arrival delays has become crucial for effective planning and mitigation. This study presents a network-aware, airline-specific framework for predicting flight delays in U.S. domestic air transportation systems using tree-based ensemble machine learning models. A large-scale dataset of 1.98 million flights, enriched with weather information, is used to develop predictive models for both departure and arrival delays. To capture the structural and operational complexity of …

Explainable Physics-Based Constraints On Reinforcement Learning For Accelerator Optimization, Jonathan Colen, Malachi Schram, Kishansingh Rajput, Armen Kasparian

Data Science Faculty Publications

We present a reinforcement learning (RL) framework for optimizing particle accelerator experiments that builds explainable physics-based constraints on agent behavior. The goal is to increase transparency and trust by letting users verify that the agent’s decision-making process incorporates suitable physics. Our algorithm uses a learnable surrogate function for physical observables, such as energy, and uses them to fine-tune how actions are chosen. This surrogate can be represented by a neural network or by an interpretable sparse dictionary model. We test our algorithm on a range of particle accelerator optimization environments designed to emulate the Continuous Electron Beam Accelerator Facility at …

Volatility Modeling With An Application To Risk Parity Portfolios, Kenneth Hou

CMC Senior Theses

This thesis studies volatility modeling in the context of risk parity portfolio construction. I compare three risk parity portfolios that differ only in their underlying volatility model: a historical covariance baseline, a Bayesian stochastic volatility model, and a GRU–GARCH hybrid neural network. Using daily returns on Kenneth French’s five industry portfolios from January 2016 through December 2025, I construct monthly rebalanced portfolios under each model, with the SV and GRU forecasts embedded in hybrid covariance matrices that combine forecasted volatilities with rolling historical correlations. The results document a divergence between forecast accuracy and portfolio performance: the SV model is the …

Differences In Biologic Clinical Trials For Chronic Rhinosinusitis With Nasal Polyps—Are We Comparing Apples With Oranges?, Marjolein Cornet, Peter W. Hellings, Martin Desrosiers, Martin Wagenmann, Richard Follows, Laura Walrave, Luz Adriana Jimenez, Lee Tombs, Dawn Edwards, Peter Howarth, Joseph K. Han

Department of Otolaryngology (ENT) Faculty Publications

In recent years, several biologics targeting Type 2 inflammation have been developed for treating chronic rhinosinusitis with nasal polyps (CRSwNP). These have been studied in registrational randomized controlled trials (RCTs), which vary in their patient populations, trial design, endpoints, geography, timing, or data-handling processes. While (in)direct treatment comparisons and meta-analyses have been carried out to compare efficacy results from RCTs, often these fail to properly account for these between-study differences. Here, we summarize the key between-study differences that can influence trial outcomes and highlight the resulting challenges faced when comparing outcomes from different Phase III RCTs of biologics in CRSwNP.

Deep Learning Approaches For Voltammetric Analysis Of Coffee, Ryan Koes

Honors Theses

This thesis investigates deep learning approaches for voltammetric analysis of brewed coffee using a low-cost electrochemical system and screen-printed electrodes (SPEs). Traditional analytical methods, such as high-performance liquid chromatography (HPLC) and gas chromatography-mass spectrometry (GC-MS), provide precise quantification of key compounds but require expensive instrumentation and specialized expertise, limiting accessibility. While SPEs offer a more accessible alternative, they yielded poor results with traditional processing; however, when combined with a neural network, the system proved more effective. In experiments with 132 coffee samples, mean errors for caffeine, CGA, and TDS predictions were 52.98 ppm, 70.48 ppm, and 0.08%, respectively. These findings …

Utilizing Computer Modeling To Optimize Electric Fields Within Xenon Time Projection Chambers, Miles Meloni

Honors Theses

XENONnT is a physics experiment designed with the goal of detecting dark matter particles. The detector is a time projection chamber; a series of charged electrodes creates an electric field, surrounding a central body filled with liquid and gaseous xenon. Photomultiplier tubes (PMTs), positioned on either end of the chamber, serve to detect light signals. We seek to minimize the root mean square of the electric field norms experienced by the PMTs. This quantity corresponds to the variance in the electric field observed by the PMTs. Establishing a consistent electric field is important to maintaining these sensitive components. The electric …

Exploring The Synergy Between Very Large Transformer And Lstm Models For Effective Medical Captioning From Videos To Text: The Impact Of Captioning In Healthcare, R. V. Aswiga, Moin Ahmed Zahir

Data Science Faculty Publications

In today’s rapidly evolving digital landscape, the demand for accurate and contextually relevant subtitles for image and video content, particularly in the medical domain, is increasingly critical. Despite the proliferation of visual data across various platforms, existing captioning systems often struggle due to variations in visual settings, complex temporal relationships, and nuanced semantics. Additionally, challenges such as limited datasets, privacy issues, and specialized annotation requirements make medical image captioning particularly difficult. To tackle these challenges, we investigate cutting-edge deep learning methodologies, specifically Transfer Learning and Transformer models, through a comparative analysis. Specifically, we focus on Transfer Learning through the MedVisionCapturer …

Advancing Hate Speech Detection: Binary And Multiclass Approaches From Traditional Methods To Parameter-Efficient And Ontology-Guided Language Models, Mahmoud Abusaqer

Graduate Theses/Dissertations

The widespread proliferation of hate speech on social media platforms poses significant challenges for content moderation and user safety, requiring automated systems that are simultaneously accurate, efficient, and capable of fine-grained distinctions. This thesis investigates hate speech detection through five published manuscripts organized into two complementary threads: binary detection (hateful vs. non-hateful) and multiclass detection across demographic targeting categories. The binary thread progresses from a broad 38-model baseline spanning traditional machine learning, deep learning, and transformer architectures (where RoBERTa reaches 91.48% accuracy and CatBoost remains competitive at 88.60%) to parameter-efficient adaptation, in which Low-Rank Adaptation (LoRA) of large language models …

Multiple Changepoint Detection For Non-Gaussian Time Series, Robert Lund, Thomas J. Fisher, Norou Diawara, Michael Wehner

Mathematics & Statistics Faculty Publications

This article combines methods from existing techniques to identify multiple changepoints in non‐Gaussian autocorrelated time series. A transformation is used to convert a Gaussian series into a non‐Gaussian series, enabling penalized likelihood methods to handle non‐Gaussian scenarios. When the marginal distribution of the data is continuous, the methods essentially reduce to the change of variables formula for probability densities. When the marginal distribution is count‐oriented, Hermite expansions and particle filtering techniques are used to quantify the scenario. Simulations demonstrating the efficacy of the methods are given and two data sets are analyzed: 1) the proportion of home runs hit by …

An Ensemble Classifier For Ordinal Outcomes In High-Dimensional Genomics Data, Heranga K. Rathnasekara, Sinjini Sikdar

Mathematics & Statistics Faculty Publications

Analysis of genomics data for predicting disease outcomes is a fast-growing field in medical research. There often exist categorical, specifically, ordinal outcomes that need to be predicted based on genomic profiles. This has led to recent development of some high-dimensional ordinal classification methods that can address the large dimensionality of the genomic covariate set. These high-dimensional ordinal models tend to vary widely in their performance depending on the data they are applied to and the evaluation criteria used. In this article, we outline an ensemble ordinal classifier that integrates different ordinal modeling approaches through bootstrap-based model evaluation, multi-metric performance assessment, …

Beyond The Lace Index: Benchmarking Machine Learning Architectures And Explaining 30-Day Hospital Readmission Risk With Shap Analysis, Carl E. Hughes Iii

Williams Honors College, Honors Research Projects

Unplanned 30-day hospital readmission remains a fundamental challenge in US healthcare, associated with increased risk to patient recovery and representing an estimated $52.4 billion in annual expenses (Beauvais et al., 2022). While the rigorously validated LACE index serves as the clinical standard for readmission modeling, its linear structure and four explanatory variables lack the complexity to capture the high-dimensional and interactive nature of patient risk. This study utilizes an admission granularity level cohort of the MIMIC-IV database to develop and compare machine learning architectures against the baseline LACE index. Due to the imbalanced prevalence of readmission, the penalized logistic regression, …

Modeling Housing Prices: Which Features Matter Most?, Alex Ruvolo

Williams Honors College, Honors Research Projects

This paper attempts to find the biggest factors and traits that influence the cost of housing. This will include the lot size, type of street, utilities, neighborhood, year built, heating, electrical, yard size, number of different rooms, age, condition, and others. I will attempt to answer the question of whether the prices of houses have changed within the last 5 to 10 years, and obviously this is an easy question to answer. However, the bigger question beyond this is are the main factors affecting housing prices all important in explaining this relationship? Is one factor more important than the rest …

Srgan-Based Deep Learning Framework For Wind Turbine Damage Detection From Sentinel-2 Imagery, Kübra Çakir, Onur Elma, Murat Kuzlu

Engineering Technology Faculty Publications

The operational reliability of wind turbines is critical for sustainable energy production in smart grids. This study proposes a remote monitoring approach using perceptually enhanced satellite imagery. Sentinel-2 multispectral data (10 m resolution) has been processed with a Super-Resolution Generative Adversarial Network (SRGAN) to improve visual quality to a perceptual resolution of 30 cm. Although true spatial refinement is not achieved, the sharper structural details enhance classification accuracy. The data set comprises 15,000 images—10,000 SRGAN-enhanced and 5000 augmented through rotation, zoom in, increasing brightness, noise addition, and blurring. A custom Convolutional Neural Network (CNN) has been trained to classify turbines …

Integer-Valued Time Series Model Via Copula-Based Bivariate Skellam Distribution, Mohammed Alqawba, Norou Diawara, Mame Mor Sene

Mathematics & Statistics Faculty Publications

Time series analysis is crucial for modeling and forecasting diverse real-world phenomena. Traditional models typically assume continuous-valued data; however, many applications involve integer-valued series, often including negative integers. This paper introduces an approach that combines copula theory with the bivariate Skellam distribution to handle such integer-valued data effectively. Copulas are widely recognized for capturing complex dependencies among variables. By integrating copulas, our proposed method respects integer constraints while modeling positive, negative, and temporal dependencies accurately. Through simulation and an empirical study on a real-life example, we demonstrate that our class of models performs well. This approach has broad applicability in …

A Bayesian Late-Fusion Supportability Framework For Rare-Disease Severity Prediction In Glut1 Deficiency Syndrome, Jordan M. Rodriguez

Mathematics Dissertations

Glucose transporter type 1 deficiency syndrome (GLUT1-DS) is a rare neurometabolic disorder with heterogeneous neurological and developmental severity. Because patient-level severity is not observed as a single validated outcome, this dissertation develops a Bayesian late-fusion supportability framework for constructing and predicting an ordered latent severity phenotype from clinical, genetic, and EEG-derived evidence. The primary target was constructed in a larger clinical cohort using age-5 symptom burden and learning cognition, then assigned to an aligned multimodal prediction cohort. Target-defining variables were excluded from supervised predictors, and models were evaluated using patient-exclusive cross-validation with training-fold preprocessing and fold-wise EEG PCA.

The primary …

A Major Update And Improved Validation Functionality In The Mwtab Python Library And The Metabolomics Workbench File Status Website, P. Travis Thompson, Hunter N. B. Moseley

Markey Cancer Center Faculty Publications

Background: The Metabolomics Workbench (MW) is a public scientific data repository consisting of experimental data and metadata from metabolomics studies collected with mass spectroscopy (MS) and nuclear magnetic resonance (NMR) analyses. Although not as rapidly as in the past, MW has steadily evolved, updating its mwTab and JSON deposition text file formats and its web-based infrastructure. However, the growth of MW has been exponential since its inception in 2013 and continues to be exponential, with the number of datasets hosted on the repository increasing by 50% since April 2024. As part of regular maintenance to keep up with changes to …