Data Science Commons^™

Addressing The Problems Of Data Variations, Quality, And Scarcity In Training Deep Neural Networks, Jian Sun

Electronic Theses and Dissertations

The performance of deep neural networks (DNNs) is strongly influenced by the characteristics and quality of the underlying datasets. This Ph.D. dissertation addresses three pervasive data challenges-imbalance, quality degradation, and scarcity-that commonly hinder the effectiveness of DNNs in computer vision (CV) and natural language processing (NLP) applications.

Class imbalance remains one of the most frequent causes of degraded model generalization. While Focal Loss effectively mitigates inter-class imbalance by assigning higher weights to minority classes, it struggles with intra-class imbalance, particularly in video datasets where longer clips dominate feature representation. To address this, I implement and utilize …

What Would It Take To Compose The Ideal Library Dashboard As A 'Symphony' Of Library Data?, Evan Rusch, Heidi J. Southworth, Nat Gustafson-Sundell

Library Services Publications

Most areas of the library produce data informative about the scope and success of library services. Many areas use that data, but there might be a variety of approaches. In our library, we have developed online, interactive dashboards to understand the value of collections on our campus. Other library service areas might produce one-shot reports encapsulating their data, or they rely on analytics silos specific to their services. Our hope is to work toward a dashboard of all or most library services, possibly to include other learning services under the library roof. Instead of a ‘battle of the bands,’ we …

Kms-Net: Kolmogorov–Arnold-Based Multi-Scale Attention Network For Cardiac Segmentation, Abid Mehmood, Hassan Ali, David Noule Tolno, Sery Gahouidi Thierry S, Muhammad Saeed, Naeem Ahmed

Research & Publications

Accurate segmentation of cardiac structures in 2D echocardiography is essential for diagnosing cardiovascular disease and computing clinical metrics such as chamber volumes and ejection fraction. Conventional U-Net architectures excel at extracting local spatial features but struggle with long-range dependencies inherent in noisy ultrasound images, while pure Transformer-based models capture global context at the expense of fine boundary detail. To address these limitations, we propose KMS-Net, a novel hybrid segmentation architecture that integrates Kolmogorov–Arnold Networks (KANs), a class of learnable, spline-based function approximators that replace fixed activation functions with trainable nonlinear mappings, alongside multi-scale attention mechanisms. Specifically, spline-based KAN layers (grid …

Terrorism By The Numbers: Event And Structural Determinants Of Attack Outcomes, Claire Lebakken

Student Research Symposium (SRS)

This project aims to identify which event-level and structural covariates are most predictive of terrorism outcomes. Using the Global Terrorism Database (1970–2020), we examine whether fatalities, injuries, attack type, target type, and actor type, combined with national-level conditions, can reliably predict outcomes such as lone actor versus group involvement, attack method, target selection, and property damage. Event-level data are merged with World Bank Development Indicators and Freedom House scores to incorporate economic and governance contexts. After cleaning the data, creating dummy variables, and log-transforming skewed measures (e.g., GDP per capita), we apply logistic and multinomial logistic regression models to test …

Associational Inference With Many Potential Covariates: Bayesian Information Criterion Elastic Net, Farideh Bagherzadeh Khiabani, I-Chan Huang, Jose Miguel Martinez Martinez, Shizue Izumi, Sedigheh Mirzaei, Tiange Zheng, Irina Dinu, Yutaka Yasui

COBRA Preprint Series

Background: An emerging feature in modern biomedical research is collecting and analyzing numerous variables. In the presence of many potential covariates, inference becomes challenging requiring both distinguishing a set of covariates truly associated with an outcome and estimating their corresponding regression coefficients consistently. Traditional statistical inference typically focuses on estimating coefficients assuming a pre-specified set of covariates. Further, advanced machine/statistical learning methods performing both selection and estimation predominantly focus on outcome prediction rather than association inference.

Methods: Motivated by our epidemiological research on long-term childhood cancer survivors, where we aimed to investigate associations between a large pool of longitudinal symptom …

Data Centers In Mountain West Markets, 2026, Cason Noll, Krish Sharma, Maisoon Faris, Olivia K. Cheche, Caitlin J. Saladino, William E. Brown Jr.

Transportation & Infrastructure

This fact sheet reports on the distribution and geographic concentration of data centers across the Mountain West states of Arizona, Colorado, Nevada, New Mexico and Utah as of March 6th, 2026. Using data from DataCenterMap, this fact sheet examines the number of data centers in each Mountain West state and further analyzes market-level  distribution, defined as cities within each state where data centers are located. The data are used to compare state totals and to rank Mountain West markets from highest to lowest based on the number of data centers operating in that area.

Molecular Quantum Particle Algorithm (Mqpa): Hybrid Quantum-Classical Learning For Molecular Property Prediction, Jessica Mcphaul, Bivin Sadler Ph.D.

SMU Data Science Review

Classical machine learning models and quantum kernel methods often struggle to capture quantum-coherent molecular features under the constraints of noisy intermediate-scale quantum (NISQ) hardware, limiting both predictive accuracy and scalability.

This paper introduces the Molecular Quantum Particle Algorithm (MQPA), a hybrid quantum–classical framework designed to achieve chemically accurate property prediction by integrating handcrafted molecular descriptors with parameterized quantum circuits. Molecular inputs, expressed as SMILES strings, are processed via RDKit and encoded through angle-based quantum gates with entangling layers in Qiskit [1]. Quantum parameters are optimized using simultaneous perturbation stochastic approximation (SPSA) [2], while classical regression layers leverage Adam [3] …

Ai-Powered Reporting For Improved Hospital Efficiency, Joel C. Laskow, Chris Papesh, Srishti Awasthi, Jacquelyn Cheun

SMU Data Science Review

This study explores the feasibility of an AI-powered chatbot for HIPAA-aligned intake of emergency room patients seeking treatment for overdose and violence. The system utilizes AWS Amplify, an encrypted EC2 instance, and a secure S3 Bucket house on Amazon Web Services. Chat functionality is powered by a multi-agentic framework operating on Anthropic’s Claude Sonnet 4. Manual evaluation and exact match testing reveal the system reliably obtains and records relevant information during intake. Future work will focus on expanding accessibility by integrating voice functionality, obtaining HIPAA compliance certifications, and incorporating the chat system into existing healthcare networks.

Bias Evaluation Of Healthcare Data With The Use Of Vbqa - A Vaers Inspired Bias Question Answer Dataset, Nolan Dulude, Renu Karthikeyan, Bivin Sadler, Faizan Javed

SMU Data Science Review

Abstract. Large Language Models (LLMs) are being used increasingly within the healthcare industry to summarize complex clinical information, but their outputs can often reflect biases inherited from their training data. In healthcare, these biases are not just technical flaws, but they can lead to distorted and false information about vaccine safety, compromise patient trust, and lead to potential harmful outcomes. This study investigates bias found in LLM-generated outputs to question-answer pairs inspired by adverse vaccine reactions using COVID-19 data from the Vaccine Adverse Event Reporting System (VAERS) from 2020–2024. We examined whether training the LLMs on a known Bias Benchmark …

Nlp Bias And African American English, Kenya Roy, Faizan Javed

SMU Data Science Review

African American English (AAE), also referred to as African American Vernacular English (AAVE), is widely used on social media, but most sentiment analysis tools are trained only on Standard American English (SAE). This mismatch can cause models to misclassify dialectal expressions—especially by labeling neutral or positive AAE as negative or toxic. These errors matter, since Natural Language Processing (NLP) systems are now central to content moderation and brand monitoring. This research will evaluate the VADER, RoBERTa, GPT-OSS, and Gemma’s handling of AAE in comparison to SAE using the TwitterAAE corpus, a public dataset of tweets with estimated AAVE usage. The …

Automating Cardiff Model Data Capture In Emergency Departments: Ambient Nlp Integration With Oracle-Cerner Fhir Systems, Simi Augustine, Marco A. Lopez, Jacquelyn Cheun, Chris Papesh

SMU Data Science Review

Violence and overdose events in Las Vegas occur at rates above the national average, with fewer than half of violent injuries reported to law enforcement [2,7]. The Cardiff Model offers a proven framework for standardized data collection and sharing between hospitals and public safety partners, yet many implementations still rely on manual entry. We propose an ambient triage pipeline integrated with Oracle-Cerner electronic health record systems to listen to nurse–patient dialogue, convert speech to text, extract Cardiff fields, and write standards-based FHIR Bundles for analytics. Using SMART on FHIR standards and Cerner Millennium APIs, the study evaluates whether ambient capture …

Anomaly Detection For Multi-System Bug Triage, Gibran Miguel Zavala Gamero, Hayoung Cheon, Mustafa Iqbal

SMU Data Science Review

Large-scale software systems produce vast volumes of logs and telemetry, making manual incident triage slow and error prone. This study presents an unsupervised anomaly detection pipeline that fuses logs, metrics, and traces through late fusion. Using Hybrid Ensemble modeling with Isolation Forest, and Long Short-Term Memory (LSTM) Deep Learning model, the system detects cross-service anomalies producing and assigning a composite triage score reflecting severity and impact. Ranked alerts are categorized into Critical, High, or Medium priorities for review. A retrieval-augmented generation (RAG) layer enriches results with contextual summaries for explainable triage. Evaluated on synthetic multi-service datasets, the pipeline …

A Modular Framework For Cost-Efficient Aspect-Based Sentiment Analysis Using Small Language Models, Senthil Kumar, Nibhrat Lohia

SMU Data Science Review

Aspect-based sentiment analysis (ABSA) links opinions in text to specific product attributes (for example, battery life, screen quality, or delivery speed) rather than only assigning an overall star rating. This level of detail is important in domains such as e-commerce, where teams need to know which features customers praised and which they criticized. Traditional ABSA pipelines have relied on large language models (LLMs), which achieved high quality but were expensive to run and difficult to scale. This study evaluated whether small language models (SLMs) in the 1–3 billion parameter range could serve as a lower-cost alternative. We implemented a modular …

Availability Model To Evaluate Ai Data Centers’ Role In Grid Stability, Troy Mcsimov, Trevor S. Kunz, Jeffrey Billo

SMU Data Science Review

The United States has made it clear; it is imperative that the US wins the global AI race. This paper focuses on one of the most challenging puzzle pieces surfaced at the POWER Data Center conference (San Antonio, Sept. 30.); for Electric Reliability Council of Texas (ERCOT) the limiting factor is not generation alone but the need to balance generation and load to preserve grid reliability.

The regulatory landscape fundamentally changed with the passage of Texas Senate Bill 6 in June 2025, which mandates new large loads must "contribute to the recovery of the interconnecting electric utility’s costs" (Texas Legislature, …

Application Of Open-Source Small Large Language Models For Finance Report Analysis, Tue Vu, Mark Austin, Marcel Tuijn

SMU Data Science Review

The rapid integration of generative AI in finance introduces both opportunities and challenges, particularly when analyzing sensitive data such as Securities and Exchange Commission (SEC) filings. This study investigates the use of open-source Small Large Language Models (SLLMs), deployed locally through the Ollama and LangChain frameworks, combined with Retrieval-Augmented Generation (RAG) for extracting financial insights relevant to index performance and reporting quality. Two key objectives guide this work: (1) benchmarking multiple open-source SLLMs for sentiment analysis, multiple-choice reasoning, and financial question answering, and (2) assessing the feasibility of locally deployed SLLMs for domain-specific financial queries. A standardized set of 50 …

Reducing Range Anxiety Through Predictive Modeling Of Ev Battery Degradation, Caleb Thornsbury, Christian Castro, Bivin Sadler

SMU Data Science Review

Electric Vehicles (EV) range anxiety remains one of the top barriers for broader adoption. Range anxiety can be attributed to battery pack age and degradation over time. This paper plans to explore how to address this issue by creating a machine learning model that can predict degradation based on usage, temperature, battery chemistry, charging habits and exploring whether other factors tie into range degradation. This research will be using real world charging data along with lab tested chemistry data to build a model that can be chemistry specific for degradation. This paper will help perspective used-EV buyers learn about battery …

“It’S A Lot More To It Than Just Research”: Integrating Critical Data Literacy And Reasoning With Data Into A Stem Summer Camp, Marc T. Sager, Saki L. Milton, Candace Walkington

Publications

Purpose: This study explores how middle-grade girls from predominantly underrepresented and underserved racially and ethnically minoritized (UUREM) backgrounds developed critical data literacy (CDL) through participation in a week-long residential STEM camp. Given the increasing importance of data science education in a data-driven world, this research examines how informal learning environments can support CDL development among youth from historically marginalized groups.

Design/Methodology/Approach: The study draws on qualitative interview data from eleven participants, and their group-produced artifacts to investigate how the camp experience supported engagement with data and the development of CDL. Interviews explored participants' experiences with data collection, organization, analysis, and …

Using Ensemble Disagreement To Stabilize Conformal Prediction Under Distribution Shift, Patrick D. Murphy

Master's Theses

Semantic segmentation of eelgrass from drone imagery is crucial for coastal habitat monitoring, restoration, and management, as these habitats continue to see rapid changes due to climate change and human influence. However, the reliability of generalizing a deployed classification model relies on both high-accuracy segmentation as well as robust uncertainty quantification that holds up when conditions change over years or locations. Conformal prediction (CP) is a method that converts a classifier's output into prediction sets with a guaranteed average coverage level for in-distribution data. However, the “vanilla” conformal score can often under-cover in hard or out-of-distribution (OOD) regions under drift. …

Student Mental Health: Screening For Stress, Anxiety And Depression Using Fitbit Data, Rebecca Lopez, Avantika Shrestha, Ml Tlachac, Kevin Hickey, Xingting Guo, Shichao Liu, Elke Rundensteiner

Information Systems and Analytics Department Faculty Conference Proceedings

College students experience many stressors, resulting in high levels of anxiety and depression. Wearable technology provides unobtrusive sensor data that can be used for the early detection of mental illness. However, current research is limited concerning the variety of psychological instruments administered, physiological modalities, and time series parameters. In this research, we collect the Student Mental and Environmental Health (StudentMEH) Fitbit dataset from students at our institution during the pandemic. We assess the ability of predictive machine learning models to screen for depression, anxiety, and stress using different Fitbit modalities. Our findings indicate potential in physiological modalities such as heart …

A Unified Methodological Framework For Generating Digital Twins Of Multi Class Uncrewed Systems (Uxs), Sai Raghava Pathuri

Shelby Hall Graduate Research Forum Presentations

No abstract provided.