Mason Journals (George Mason Univ.)
Not a member yet
3256 research outputs found
Sort by
A Direct Survey-Based Synthetic Population Generation Approach for Small Area Estimates
No full textSmall area estimation (SAE) is a statistical technique that can be used to analyze health outcomes and characteristics in small geographic areas. While the Centers for Disease Control and Prevention (CDC) sets a widely accepted standard for SAE using a multilevel regression approach, its methods are complex and not readily available for public use. To address this gap, we propose an open and more accessible approach that directly integrates public health survey data into the widely used iterative proportional fitting (IPF) technique to generate small area estimates for public health applications. We apply this method to estimate cancer, diabetes, and chronic obstructive pulmonary disease (COPD) prevalence, incorporating demographic factors such as age, race, income, gender, and education, along with health indicators including smoking status, body mass index, health insurance coverage, and urban living status. Our cancer estimates, when compared with New York county-level data, achieved an R2 of 0.555, comparable to CDC’s estimates, with an R2 of 0.61. For diabetes, comparison with Florida county-level data yielded an R2 of 0.437 (CDC: 0.475). For COPD, our estimates achieved an R2 of 0.462, surpassing the CDC with an R2 of 0.426. These results demonstrate that our approach can reasonably generate SAEs that are as accurate or more accurate than the gold standard by CDC. This study contributes to advancing SAE by offering an open and publicly available alternative for generating estimates that does not require complex statistical expertise, expanding access to tools that support public health research and decision-making
Social Media User Characteristics of US Counties
No full textHuman-deer interactions are increasing due to human urbanization into natural habitats, which has raised concerns about the transmission of zoonotic diseases. Social media videos can be used as a data source to see how these interactions play out in geographical regions within the United States. Since social media usage can be biased across platforms, it's important to understand the demographic determinants of platform usage. Therefore, this study investigates the demographic characteristics associated with social media usage to assess how user bases vary across platforms. We collected demographic variables and the estimated number of adults that have used TikTok, Instagram, and Facebook for each US county from Esri Business Analyst, removed highly correlated variables such as pet ownership, standardized the dataset, and conducted Ordinary Least Squares (OLS) regression analyses. The OLS models show that education and income are negatively associated with Facebook and TikTok usage, and that TikTok users are significantly younger. Instagram usage, in contrast, is positively associated with higher education, higher income, urban populations, and Democratic affiliation, contrasting Facebook users. The variable "male" is negatively associated with all three platforms, indicating higher female usage. TikTok usage, followed by Instagram, increases most in areas with Hispanic populations, while Facebook usage increases most in areas with larger Black/African American populations. Instagram usage, however, decreases in those areas. The models explain 71% (Instagram), 75% (Facebook), and 91% (TikTok) of the variation in platform usage. Identifying the demographic characteristics of social media platforms mitigate bias common in social media analysis of wildlife-human interactions.  
Leveraging the Table of Contents to Improve Efficiency in an RAG-Powered Chatbot
No full textSectors have increasingly demanded document processing, drawing attention to Retrieval-Augmented Generation (RAG) systems, which leverage large language models and document search to provide answers exclusively from a specific knowledge base. However, RAG with large documents can be inefficient, with weak points such as high latency and memory usage. This research aims to use the Table of Contents(TOC) to narrow the search space. The techniques examined are Keyword-First Pre-filtering and TOC Query Routing. Keyword-First Pre-filtering uses query-TOC keyword matching to cut out irrelevant parts of the document, dramatically reducing unnecessary computation. If keywords do not exactly match, this technique can fall back on semantic matching. Conversely, TOC Query Routing involves processing the whole document but using the TOC to dynamically guide focus to certain sections. To test impact, 27 natural language queries relating to the DAFMAN 36-2664 policy document, created by Lt. Col. John McKee, were fed into each model. Accuracy, average latency, and average memory usage across all models were recorded. The following figures are derived from a comparison with a near-identical model that used brute-force RAG. The base model used Apache Tika for text extraction, then parsed and chunked all the information. These chunks were embedded through the all-MiniLM-L6-v2 sentence-transformer model and stored with ChromaDB. Using Meta's LLaMA 3 8B model, the base model was then able to generate context-aware answers. It was found that Keyword-First Pre-filtering could reduce latency by 54% and memory usage by 87%, while maintaining answer accuracy. TOC Query Routing decreased latency by 32% while maintaining answer accuracy. These findings suggest that TOC-driven strategies can significantly improve the efficiency of RAG systems without compromising accuracy(with Keyword-First Pre-filtering being especially promising), making them ideal for environments like the Department of the Air Force, where speed and resource constraints are critical. This research could be expanded upon by exploring how search spaces can be limited effectively for queries that involve chunks from disparate areas in the document
Taxonomic Labeling of Open Ended Feedback: Identifying Cognitive Presence in Survey of STEM Graduate Students
No full textCognitive presence is the extent to which learners are able to construct meaning through discourse and reflection. Evidence of cognitive presence in students indicates the cultivation of a purposeful learning community, a use of critical thinking skills, and a deeper level of understanding of course material. A series of inquiry-based research learning modules was designed to facilitate cognitive presence in STEM graduate students. Following the completion of research learning modules, students responded to a survey that included three free-response questions regarding elements to keep, elements disliked, and elements to add to the course. This study aims to analyze students’ responses to the three free-response questions for evidence of cognitive presence. Student feedback was organized by two student researchers using a hierarchical taxonomy system. Content-focused responses (WHAT categories) used three levels: main categories, categories, and sub-categories. Cognitive presence (WHY categories) used a two-tiered system of main categories and categories. The taxonomy facilitated data labeling of over 300 open-ended student responses. Preliminary analysis of the WHAT categories identified course components students valued. Practical Applications, Topic Exploration, and Peer Review were among the most frequently identified components. WHY categories revealed cognitive presence indicators, with “improving understanding” and “interesting” content being the primary indicators. Future work can include a comprehensive statistical analysis of categorized feedback to better identify patterns across student demographics (online vs. in-person, domestic vs. international) and course components. Statistical analysis will examine the correlations between categories to develop recommendations for optimizing inquiry-based course design
Personalized Debugging Feedback: A Comparative Study of AI Assistants for Novice and Expert Programming Tasks on CodeForces
No full textCode debugging is an essential skill for a student programmer that must be applied efficiently and effectively. However, teaching debugging techniques in a classroom setting is sporadic [1]. Although Artificial Intelligence (AI) has been widely integrated into various computer science tasks, no analysis was found on its effectiveness in debugging code for varying levels of programming expertise. After conducting a literature review, ChatGPT, Github Copilot, Phind, and Amazon Q were selected as notable AI debuggers. This study evaluated the efficiency and effectiveness of these 4 AI models on debugging code submissions for 18 CodeForces tasks, ranging from novice (800 rating) to advanced (1500 rating). The code simulated by each AI model was evaluated by four student researchers on a 500-point rubric on their accuracy (200 points), comprehensibility (100 points), code explanation quality (100 points), and similarity of code logic to the user’s inputted solution (100 points), and the average was used for analysis. The results indicate that GitHub CoPilot and ChatGPT displayed the highest overall effectiveness for novice debugging tasks, with an average score of 455.55 out of 500 points (91.1%), and Amazon Q exceeded other AI models for advanced debugging tasks, with an average score of 444.44 out of 500 points (88.9%). Overall, ChatGPT had the highest average score across both novice and advanced programming questions, with an average score of 438.89 out of 500 points (88.0%) on the 18 coding tasks. These findings demonstrate the strong potential for AI models to assist with debugging, specifically, for novice programming tasks. Future research would involve expanding the dataset to include various programming languages and levels of expertise, enabling the development of enhanced systems for bug resolution.
[1] Noller, Y., Chandra, E., HC, S., Choo, K., Jegourel, C., Kurniawan, O., & Poskitt, C. M. (2025), Simulated Interactive Debugging, arXiv, https://arxiv.org/abs/2501.0969
Design and Characterization of a Coumarin-Based Hydrazine Fluorescent Probe for Labeling Biomolecular Carbonyls and Detecting Oxidative Stress in Live Cells
No full textOxidative stress—an imbalance between reactive oxygen species (ROS) and the body's antioxidant defenses—is a key contributor to diseases such as cancer, fibrosis, and neurodegeneration. To study and treat these diseases, it is essential to identify oxidative damage in tissues with tools that offer high spatial resolution, rapid detection and minimal disruption of the existing environment. However, achieving this level of detail is challenging with current technologies, such as PET and MRI scans, which either rely on ionizing radiation or have low sensitivity, respectively. Fluorescent probes offer a safer, higher-resolution, real-time imaging alternative, but many current designs are non-selective and non-specific in their labeling. To address these gaps, we designed, synthesized, and characterized a coumarin-based hydrazine fluorescent probe that forms hydrazones with carbonyls like allysine, a key biomarker of oxidative stress, using bioorthogonal click-like chemistry. Prior research in our lab demonstrated that strategically placing electron-withdrawing and electron-donating groups, along with changing π-conjugation, influences fluorescence properties and shifts absorbance wavelengths—insights that guided the structural design of this probe. Synthesis of the probe was completed in three steps and was characterized using 1H-NMR, 13C-NMR, UV-Vis, and fluorescence spectroscopy. A kinetic assay was conducted to establish the probe’s reactivity and rate of reaction, revealing second-order behavior. This work presents a non-invasive, real-time fluorescent probe for the early detection of oxidative stress, enabling precise disease diagnostics and real-time tracking of therapeutic responses
Life cycle analysis and environmental impact of the lithium-ion NMC 811 battery
No full textThe lithium-ion NMC 811 battery, composed of 80% nickel, 10% manganese, and 10% cobalt in weight, is a viable option for use in electric vehicles (EVs) due to its high energy density, long lifespan, and reduced cobalt content. As EVs are becoming increasingly popular, understanding the environmental impact of NMC 811 cells through life cycle analysis (LCA) is crucial in assessing their sustainability for more long-term and widespread use. Using the LCA software OpenLCA and analyzing factors such as the resources and processes needed to produce NMC cells, more information can be discovered regarding how much each specific component of NMC battery production impacts the environment. A simulation is run representing the raw material extraction, manufacturing, transportation, use, and recycling phases of NMC 811. The various differing quantities of resources account for various percentages of total energy consumption, and some processes within the production of batteries consume much larger amounts of energy. Results of the simulation would show the energy consumption distribution and environment impact contribution of the NMC 811 battery [1]. A greater understanding of the life cycle of NMC can help reduce environmental impact with additional research. Knowing which processes account for the most energy consumption can be helpful in developing more sustainable methods as they can be targeted and improved upon.
[1] Y. Deng, J. Li, T. Li, X. Gao, C. Yuan, Journal of Power Sources, 2017, 343, 284