|
Maine Marsh Management: A Survey of Maine Property Owners and their Views on Shoreline Management
Dr. Caroline Noblet (UMaine), CJ Evans-Ralston (UD)*, Melissa Godin (UMaine), Dr. Kent Messer (UMaine).
Despite their ecological and economic value, salt marshes along the Atlantic Coast face severe threats from sea level rise, climate change, and encroaching coastal development (Burman et al., 2024). A significant barrier to conservation is the decentralized decision making of private coastal landowners. Some of these decisions, such as grey shoreline “armoring”, have been linked to reduction in ecosystem services (Gittman et al., 2016), habitat loss (Scyphers et al., 2020), and damage to neighboring beaches (Scyphers et al., 2015). By understanding the unique needs and motivations of landowners, conservationists can design more inclusive and effective programs for salt marsh preservation.
We surveyed 862 Maine landowners in a salt marsh corridor. The survey captured information on perceptions of coastal change, outcomes for their property from sea-level rise, benefits and concerns associated with marsh on their property, and their perceptions of actions being undertaken by their neighbors related to coastal conservation or armament. An embedded experiment within the survey tests the impact of messaging on perceptions of marshlands and preferences for engaging in conservation or other actions or their land. Initial analysis indicates [CN1.1]the messaging treatment highlighting the salt marshes’ storm protection properties influenced landowners to select “flood protection” as a benefit of salt marshes (p-value = 0.001). Participants have a positive outlook of salt marshes (46% of respondents had no concerns related to salt marshes). 76% of respondents were willing to participate in at least one of the actions we list. As such, landowners are interested in taking a variety of actions for their land, though respondents who have owned their land longer prefer actions which do not alter their property. Ordered probit analysis reveals an “offsetting” effect, where participants may choose to participate in an action they view as ecologically friendly when they view their neighbors as likely to armor.
|
|
Computational Approaches to Art Conservation in Winterthur Museum
Asritha Polu*, Kevin (University Of Delaware ,) Liora(Wintherthur Museum ),Rossie,Mellisa(Winterthur Museum)
This poster presents two independent data science projects conducted in collaboration with Winterthur Museum, both applying data-intensive computational methods to support cultural heritage preservation. Project 1 develops an automated pipeline for X-ray fluorescence (XRF) spectral analysis of 19th-century book bindings, screening 511 books for toxic pigments containing lead (Pb), arsenic (As), mercury (Hg), and other hazardous elements. Using signal-to-noise ratio analysis with dynamic Rh-calibrated thresholding and spectral overlap resolution, we identified that 46.6% of the analyzed books contain detectable levels of lead, enabling targeted safety protocols for handlers and researchers. Project 2 analyzes over 10 years of longitudinal HVAC sensor data and high-resolution QuantAQ particulate data to characterize environmental conditions within Winterthur’s museum facilities, identifying temperature fluctuation events and particle size distributions that inform preservation strategies. Both projects demonstrate the power of Python-based data science tools—including Pandas, NumPy, SciPy, Ipywidgets, and Plotly—to transform raw sensor and spectroscopy data into actionable conservation insights.
|
|
Modeling Pitch-Level Impact in Baseball Using Change in Expected Runs
Suma Tata (UD)*, Abhin Suresh (UD), Tanner Elliott (UD), Jack Davis (UD)
This project develops a pitch-level model to measure how individual pitch attributes influence scoring in baseball. Rather than relying only on traditional run expectancy tables, we model the change in expected runs following each pitch as a continuous outcome. The dataset consists of more than 3.5 million collegiate pitches from over 12,900 games across the 2024 and 2025 seasons. For each pitch, base-out-ball-strike states are reconstructed and global run expectancy values are calculated to define the target variable as the difference in expected runs between consecutive pitches.
Pitch-level physical attributes such as velocity, spin rate, movement, and release metrics are used as predictors, along with engineered differential features defined relative to each pitcher’s fastest pitch type. A supervised regression framework is applied to predict change in expected runs, and SHAP-based analysis is used to examine which pitch characteristics most strongly influence run value, as well as how they interact with each other. By focusing on expected run differentials rather than specific outcomes, such as swing and miss likelihood, this approach evaluates the marginal impact of pitches in a more direct way.
In addition to model development, the project includes an applied analytics component. Processed outputs are integrated into interactive visualization of expected run states, pitch impact distributions, and player performance trends. A longer-term goal is to automate data ingestion and transition to an Azure-hosted database architecture to support scalable deployment.
|
|
Topology Optimization and Additive Manufacturing of Continuous Carbon Fiber Thermoset Composite
Md Habib Ullah Khan (UD)*, Md Mohaiminul Islam (TU), Kaiyue Deng (UD), Ismail Mujtaba Khan (UD), Ling Liu (TU), Kelvin Fu (UD)
Continuous carbon fiber composites exhibit outstanding strength-to-weight ratios, but their structural potential is often underutilized due to the limitations of conventional layup and molding methods. A key barrier is the difficulty of fabricating complex geometries while maintaining continuous fiber alignment along load-bearing trajectories. To address this challenge, we establish a design-to-manufacturing framework that integrates topology optimization, fiber path planning, and additive manufacturing for the fabrication of composite structures.The design stage employs a two-step topology optimization strategy. First, a Solid Isotropic Material with Penalization (SIMP) formulation identifies the optimal material distribution under applied loads and boundary conditions. This continuum design is then reduced to a beam-element representation, which refines anisotropic member thicknesses and connectivity for manufacturability. To ensure continuous fiber placement, a fiber path planning algorithm is introduced, generating feasible trajectories that follow the optimized truss network while avoiding dead-ends, entanglement, and deposition collisions. The optimized design was physically realized through an additive manufacturing process that spatially deposits continuous single-tow carbon fibers with more than 90% material savings. This study demonstrates the synergy of computational design and fabrication by linking SIMP-based optimization, beam-level refinement, and path planning directly to manufacturing. The approach provides a scalable route to topology-optimized continuous fiber composites with unprecedented geometric freedom, structural efficiency, and material utilization. Beyond aerospace brackets, this methodology can be extended to crashworthy automotive components, drone structures, and lightweight infrastructure systems, representing a new paradigm for high-performance composite design.
|
|
Human-Hydrologic Feedbacks and Salinization of Agricultural Land in Delaware
Hazel Helm (UD)*, Max Phelps (UD), Dr. Ancilla Inocencio (UD), Dr. Leah Palm-Forster (UD)
The Mid-Atlantic is experiencing sea level rise at rates more than twice the global average, putting Delaware at risk for saltwater intrusion and the resulting salt patches caused by high levels of surficial salt. High levels of soil salts prevents most plant species from growing, including common agricultural crops. With 43% of Delaware’s land being used for agriculture as of 2019, and with 68% of farmland within 200m of a visible salt patch between 2011 and 2017, understanding the relationship between agriculture and saltwater intrusion is key (USDA, 2019) (Mondal, 2023). While visible salt patches more frequently appear at lower elevations closer to the coast, 45% of visible salt patches in 2017 occurred at elevations of 10m above sea level or higher, an elevation range where 68% of agricultural land in the state can be found. In this project, we investigate potential factors affecting soil salinity at higher elevations, including water withdrawals for irrigation and proximity to tidal waterways. We examine relationships between these factors and the presence of salt patches on agricultural lands. Additionally, we study the likely geographic scale and economic impacts of salinization on Delaware agriculture.
Sources
Mondal, P., Walter, M., Miller, J. et al. The spread and cost of saltwater intrusion in the US Mid-Atlantic. Nat Sustain 6, 1352–1362 (2023). https://doi.org/10.1038/s41893-023-01186-6.
National Agricultural Statistics Service, United states Development of Agriculture (USDA) Delaware field office (2019), https://www.nass.usda.gov/Statistics_by_State/Delaware/About_Us/index.php.
|
|
Land use change on coastal farmland in Delaware
Max Phelps (UD)*, Hazel Helm (UD), Ancilla Inocencio (UD), Leah Palm-Forster (UD)
Salt marshes are migrating inland due to sea level rise (SLR). Protecting native salt marsh habitat provides a variety of ecosystem services including protection from storm surges and enhanced water quality. Previous studies have identified coastal forests and upland agricultural land as vulnerable land covers for marsh transgression, and significant land use changes have been observed on forest-marsh fringes and coastal farmland in Delaware across the past 30 years. Once saltwater intrusion (SWI) impacts coastal forest and farmland, native trees and crops are unable to grow in the salty soil. Intensive and long-term management actions are needed to help transition land into native salt marsh habitat, but there are many barriers to enacting management actions that are not well understood. One example is coastal farmlands enrolled in Delaware’s Agland Preservation Program (APP). APP places an easement on farmland which ensures it is kept in farmland in perpetuity. However, there are an increasing number of farmland parcels in APP that are likely to become untenable and abandoned due to SWI. The fate of these farms is unknown as enrollment in APP may prevent landowners from taking steps to convert abandoned farmland into native salt marsh habitat or another land use. To better understand the extent of this issue in Delaware, I examine the intersection of agricultural parcels enrolled in the APP and parcels that have been identified as having salt patches or being at risk for salt marsh migration. Identifying where preserved farmland intersects with areas vulnerable to saltwater intrusion can inform strategies that support both coastal resilience and ecosystem service provision. These findings can help guide proactive planning for marsh migration while addressing the future of vulnerable agricultural lands.
|
|
Fractional Optimization for learning systems
Jing Li (DSU)
Optimization plays a central role in modern learning systems, where the training dynamics largely determine convergence behavior, stability, and generalization performance. Conventional optimization algorithms, such as gradient descent and adaptive methods, are typically based on integer-order calculus and therefore rely primarily on local gradient information with short-term memory. However, many complex learning processes, particularly those involving time series, dynamical systems, and large-scale neural networks, exhibit long-range dependencies and nonlocal behaviors that are not fully captured by traditional optimization frameworks.In this work, we propose a fractional optimization framework for learning systems that incorporates fractional-order dynamics into the training process. By replacing the conventional integer-order gradient dynamics with fractional-order operators, the proposed approach introduces power-law memory effects that enable the optimizer to account for historical information over extended temporal horizons. This nonlocal memory mechanism provides a principled way to smooth optimization trajectories, suppress oscillations, and improve robustness in noisy or highly nonconvex learning landscapes. The proposed framework is general and can be integrated with common learning architectures and training pipelines. We demonstrate its effectiveness through experiments on representative learning tasks, including time-series modeling using stacked recurrent neural networks and parameter estimation in dynamical systems. Comparative studies with standard optimization methods show that fractional-order optimization can improve training stability and produce smoother convergence behavior while maintaining competitive predictive performance.Overall, this study highlights the potential of fractional calculus as a mathematical foundation for enhancing learning dynamics. The proposed fractional optimization framework opens new opportunities for incorporating long-memory effects into machine learning and provides a promising direction for future research in optimization methods for complex learning systems.
|
|
Predicting College Readiness Using Machine Learning on Longitudinal Student Data
Reshmitha Uppala (UD)* Henry May (UD), Sunita Chandrasekharan (UD) Kevin Bhimani (UD), Gail Headley (UD), Ivan Gradjansky (UD), Jeffrey Klein (UD), Samuel Van Horne1 (UD)
Many education systems struggle to monitor which students are on-track to be college-ready by the time they graduate high school. Common indicators such as standardized test scores or cumulative GPA often provide only a partial view of a student’s preparedness. They do not fully capture the complex, multi-year patterns in academic performance, attendance, and socioeconomic context that shape postsecondary outcomes. As a result, schools may miss critical opportunities to provide timely, targeted support, particularly for students who would benefit most from early intervention. Furthermore, although states collect extensive longitudinal education data, this information is rarely translated into meaningful, individualized insights that directly inform career and college planning.
To address this challenge, we are developing an AI-driven predictive and decision-support system using Delaware’s K-16 statewide longitudinal student data system. Our approach involves training machine learning models, including neural networks, on historical student cohorts to identify patterns associated with outcomes such as Immediate Higher Education (IHE) enrollment and broader college and career readiness indicators. By integrating multi-year student-level features like course rigor (e.g., regular, honors, AP, etc.), course grades, standardized test scores, attendance and discipline records, and demographic variables, the models are designed to capture nonlinear relationships and complex interactions within the data. To promote transparency and responsible AI use, we incorporate explainability techniques such as SHAP to highlight the key factors influencing each prediction.
The long-term vision of this project is to develop an AI-powered, chatbot-style interface that delivers personalized career trajectory insights. When provided with a student’s academic profile, the system will retrieve relevant features, apply the trained predictive models, and generate an interpretable, data-grounded trajectory summary. This scalable framework aims to transform longitudinal education data into actionable, student-centered guidance, supporting earlier interventions, more informed decision-making, and more equitable postsecondary outcomes.
|
|
From Maize to Common Bean: Identifying Candidate Cross-Stress Transcription Factors
Rita Hayford (DSU)*, Antonette Todd (DSU), Chase Stratton (DSU)
Plants frequently experience multiple environmental stresses simultaneously, including abiotic stresses such as drought, heat, and salinity, as well as biotic stresses caused by pathogens and pests. Understanding the regulatory mechanisms that enable plants to respond to diverse stresses is critical for developing resilient crop varieties. In a recent maize study, functional annotation and meta-analysis of publicly available transcriptomic datasets identified genes involved in both biotic and abiotic stress responses (Hayford et al., 2024). This analysis revealed gene co-expression modules and gene regulatory network (GRN) hub genes responsive across multiple stress conditions. Building on these findings, we focus on a subset of 31 transcription factors (TFs) identified from the maize cross-stress co-expression network that consistently respond to both biotic and abiotic stresses. Because transcription factors play central roles in regulating gene expression, these TFs represent promising candidates for conserved stress-response regulators across plant species. In this study, we investigate whether these maize-derived TFs have potential orthologs in common bean (Phaseolus vulgaris), an important legume crop. Amino acid sequences corresponding to the maize TFs were retrieved and curated to obtain representative protein isoforms. These sequences will be used to identify candidate homologs in common bean through protein similarity searches. Candidate genes will be prioritized based on sequence similarity, conservation of transcription factor families, and domain architecture. The resulting candidate set will guide future experimental validation and transcriptomic analyses of the common bean under multiple stress conditions. This work represents an initial step toward translating cross-stress regulatory insights from maize to legumes and may contribute to a better understanding of plant stress tolerance and to the development of crops with enhanced resilience to diverse environmental challenges.
|
|
Advancing AI Literacy for the Next Generation STEM Workforce Using Cross Campus Collaboration and Ethical Frameworks
Russell Michalak (Goldey-Beacom College)* and Deanna Merritt (Goldey-Beacom College)*
Goldey-Beacom College’s Ethical AI Literacy and Critical Writing Program demonstrates how cross-campus collaboration and ethical frameworks can prepare the next-generation STEM workforce for an AI-accelerated world. Responding to rapidly evolving workforce expectations, the program ensures that graduates develop not only technical fluency with AI tools but also the ability to use them ethically, critically, and with human judgment at the center. Employers increasingly expect candidates to demonstrate responsible AI use, critical evaluation, and strong communication skills—competencies that align directly with this program’s goals.
To meet these needs, the College implements a library-led, curriculum-embedded model that ensures equitable access to AI-enhanced writing and research tools. Institution wide licensing eliminates financial barriers for first generation, multilingual, and underrepresented students, ensuring that every learner can fully participate. Embedded instruction then teaches students how to evaluate, refine, and, when necessary, reject AI generated suggestions. Program data confirms that instruction—not access alone—drives learning gains. Sections with tool access but without library-led instruction showed significantly lower improvement in writing quality, research literacy, and critical evaluation, while guided instruction consistently produced stronger, more intentional use of technology.
The program’s impact is strengthened by its collaborative infrastructure. Hirons Library leads instructional design and assessment; writing faculty integrate AI supported assignments that reinforce rhetorical and analytical outcomes; Academic Support Services provides tutoring and non AI pathways to ensure fairness; the Center for Faculty Development builds faculty capacity through tailored workshops and resources; and Career Services translates employer expectations into résumé preparation and interview coaching. This leadership from the middle model ensures coherent, equitable student experiences across the institution.
Outcomes from 179 first year students show high engagement, measurable improvements in writing, and strengthened research literacy. Student reflections further highlight long term transfer into workplace communication, graduate study, and ethical decision making—demonstrating a scalable, equity driven model for advancing AI literacy across STEM aligned pathways.
|
|
A Computational Framework for the Design of Peptide Binders and Antibody CDRs Targeting Specific Protein Surfaces
Md Mehedi Hasan (DSU)*, Shahidul M. Islam (DSU)
Designing molecules that can bind selectively to specific regions of proteins is a central challenge in therapeutic development. Peptides and antibodies are widely used as binding molecules, but identifying sequences that interact strongly with a desired protein site remains difficult because of the enormous number of possible sequence combinations. Here, we present a computational framework for designing protein-binding sequences directed toward predefined target sites. Our approach uses two complementary strategies. In the first, residues within a peptide or binding loop are optimized sequentially. At each position, possible amino acids are tested, and the most favorable residue is selected based on its predicted binding strength to the target protein, evaluated using molecular dynamics simulations and binding free energy calculations. In the second strategy, binding sequences are constructed through fragment-based growth, where short sequence segments are progressively extended across the target interface while maintaining compatibility with the protein surface. Candidate residues are similarly evaluated based on their predicted binding interactions. To demonstrate the general applicability of this framework, we applied it to the design of peptide binders and the optimization of nanobody binding loops targeting mesothelin (MSLN), a protein overexpressed in several cancers including triple-negative breast cancer (TNBC). The designed sequences show stable interactions with the target region and improved binding characteristics compared with the initial sequences. This framework provides a flexible computational strategy for designing protein-binding molecules and may facilitate the development of peptide and antibody-based therapeutics targeting diverse proteins.
|
|
Experiential Learning in Civil Engineering Technology: Integrating Real Land Development Projects into the Classroom
Dr. YuanChi Liu (Delaware Technical Community College)*
Engineering graduates often face challenges transitioning from classroom learning to real-world infrastructure design. This project explores an experiential learning approach that integrates real land development projects into Civil Engineering Technology (CET) courses to strengthen applied learning and workforce readiness.
Students work with actual site data and perform tasks similar to professional engineering practice, including site feasibility analysis, conceptual land development planning, and infrastructure layout. Using tools such as Civil 3D, GIS mapping, and hydrologic analysis software, students evaluate parcels, consider zoning and environmental constraints, and develop preliminary design concepts.
Initial outcomes indicate that integrating real-world projects improves student engagement, spatial reasoning, and understanding of engineering systems. Students also gain greater confidence in applying technical skills to practical design problems.
This approach demonstrates how experiential learning can better prepare engineering technology students for careers in infrastructure planning, land development, and civil engineering practice while strengthening the workforce pipeline for regional infrastructure development
|
|
AI Persona–Supported Library Marketing: A Student-Led Experiential Learning Model for Sustainable Digital Communication
Kloe Waymel (GBC)*, Russell Michalak (GBC)
Academic libraries increasingly rely on digital communication—such as blogs, email campaigns, and web content—to connect students with resources and learning opportunities. However, maintaining consistent digital outreach can be difficult for small institutions with limited staff. This poster presents a case study from Goldey-Beacom College demonstrating how an artificial intelligence (AI) persona–supported workflow enabled a student-led digital marketing initiative that both restored library communication capacity and created an experiential learning opportunity.
The project paired a digital marketing student worker with the library director to rebuild a stalled content pipeline using custom AI personas configured within the library’s ChatGPT environment. These personas were designed to store institutional context, model communication tone, and guide drafting and revision while maintaining human authorship and librarian oversight. Rather than generating content independently, the AI served as a structured writing support tool that helped the student brainstorm topics, develop outlines, refine drafts, and ensure consistency in voice and messaging.
AI Persona Supported Library Ma…
The resulting workflow integrated human creativity, AI-supported scaffolding, and mentorship. The student worker identified topics based on campus needs, drafted blog posts and marketing emails, revised them with AI feedback, and finalized the content after librarian review. Content was distributed through targeted email campaigns and the library website, allowing the library to rebuild its digital outreach strategy while aligning messaging with institutional branding.
This model demonstrates how AI can function as pedagogical infrastructure rather than as a replacement for human work. The workflow strengthened the student’s professional communication skills, built AI literacy, and connected marketing theory with real-world practice. For small academic libraries and other resource-constrained organizations, the case illustrates a replicable framework for combining AI tools, experiential learning, and institutional communication strategies to create sustainable and student-centered digital engagement.
|
|
Toward Standardized AI Tool Assessment: Interpreting Vendor Usage Traces in a Librarian-Led First-Year Writing Program
Adriel Raman (GBC)*, Russell Michalak (GBC)
As colleges adopt artificial intelligence (AI) tools for teaching and learning, they also face a practical challenge: how to assess whether these tools are being used in meaningful ways. This poster examines student engagement with Grammarly in a librarian-led AI literacy program embedded in a two-semester first-year writing sequence at Goldey-Beacom College. The program was developed in partnership with composition faculty and taught students to use a human-centered workflow: students created drafts, used AI-supported revision tools, and then reviewed suggestions critically before submitting final work.
Using de-identified Grammarly administrative data from Fall 2024 and Spring 2025, the study analyzed activity among active users through three vendor-defined measures: session count, days active, and percentage of sessions improved. These metrics were examined by semester, gender, and cumulative GPA. Rather than treating vendor analytics as evidence of learning or writing quality, the study interprets them as traces of how students interacted with AI-supported revision in a shared instructional setting.
Overall, engagement intensity, measured through sessions and days active, was broadly similar across GPA levels, with modest variation appearing mainly among male students in Spring 2025. In contrast, revision-flagged activity, represented by Grammarly’s percentage of sessions improved, appeared more sensitive to instructional context. A positive GPA pattern among female students in Fall 2024 did not remain evident in Spring 2025 as the shared AI literacy framework became more established.
This project offers a privacy-preserving, library-led approach to learning analytics and highlights the need for common interpretive standards for vendor-generated AI metrics. For institutions integrating AI tools into instruction, the findings suggest that usage dashboards should be read cautiously and in context, not as direct measures of student ability or learning.
|
|
Curating Institutional Memory in Academic Libraries: AI, Metadata, and Student-Led Digital Archiving & Storytelling
Rachel Gandhi (GBC)*, Russell Michalak (GBC)
Academic libraries play an important role in preserving institutional memory, yet organizing and describing archival materials is often time-intensive and resource constrained. This poster presents a student-led digital archiving initiative at Goldey-Beacom College that combined artificial intelligence (AI), metadata creation, and experiential learning to curate a visual history of the institution.
Two international undergraduate student workers collaborated with the library to digitize and organize historical silver gelatin photographs drawn from 403 SharePoint folders and 32 physical photo albums. Over 300 hours, students scanned photographs, created metadata descriptions, and organized files using project management workflows in ClickUp. AI tools were used to assist with metadata generation, helping students draft descriptive tags and keywords that improved the discoverability of archival images within the library’s digital asset management system.
Students then reviewed and refined AI-generated metadata to ensure historical accuracy, correct misidentifications, and address potential biases in automated descriptions. Their work resulted in a curated collection of 250 historically significant photographs that will appear in a visual history book of the college (1950–1990) scheduled for publication in 2026 through Arcadia Publishing’s College Campus Series.
The project also served as an experiential learning opportunity. Through reflective journaling and collaborative decision-making, students documented their evolving understanding of digital preservation, metadata ethics, and AI-assisted workflows. They also presented a SWOT analysis of AI-supported metadata generation to the institution’s academic affairs leadership, evaluating both efficiency gains and ethical considerations.
The findings demonstrate that AI can support—but not replace—human expertise in archival curation. Student participation proved essential for contextual interpretation, inclusive metadata creation, and institutional storytelling. This project offers a scalable model for academic libraries seeking to combine AI tools, student workforce development, and digital preservation initiatives to sustain institutional memory while preparing students with practical digital archiving skills.
|
|
Understanding Institutional Factors Influencing Alumni Giving Rates
Supraja Krovvidi (GBC)*
This study examines the factors influencing alumni giving rates using the Alumni Giving Data dataset, which includes data from 48 colleges. The dataset contains variables such as graduation rates, alumni giving rates, percentage of classes with fewer than 20 students, and student-faculty ratios. Using descriptive analysis, the study explores how these institutional characteristics relate to alumni engagement and financial contributions. The goal is to identify patterns that help universities better understand what drives alumni giving and how institutional policies influence long-term support from graduates.
The descriptive analysis shows that the average graduation rate across the colleges is about 83%, with most schools performing near this level. Alumni giving rates vary much more widely, averaging around 29% but ranging from as low as 7% to as high as 67%. Class sizes are generally moderate, with about 56% of classes having fewer than 20 students, and the average student-faculty ratio is roughly 11.5 students per faculty member. Correlation analysis highlights several clear relationships: graduation rate has a strong positive relationship with alumni giving, while the student-faculty ratio is negatively related to both graduation rates and giving. This suggests that institutions where students are more likely to graduate and have better access to faculty tend to see stronger alumni support. To further examine these relationships, several regression models were developed. A simple linear regression shows that higher graduation rates are associated with higher alumni giving rates and explains about 57% of the variation in giving. A multiple regression model improves the explanation to about 70%, identifying graduation rate as a positive factor and student-faculty ratio as a negative factor affecting alumni giving. The best-performing model is a polynomial regression, which explains about 75% of the variation and indicates that the relationship between graduation rates and alumni giving is not purely linear. Overall, the findings suggest that improving student success and maintaining smaller student-faculty ratios may help universities strengthen alumni engagement and increase future giving.
|
|
What Determines Home Value?
Jillian Yang (Goldey-Beacom College)*
Residential home values are shaped by a combination of local market context, physical property characteristics, and broader price trends. This study integrates descriptive analytics, regression modeling, and time-series forecasting to examine how these factors influence housing prices across multiple U.S. datasets.
First, a comparative market analysis of Athens, Georgia, and Chapel Hill, North Carolina revealed substantial geographic differences in housing value. Chapel Hill homes had a mean sale price of $429,152, compared with $219,671 in Athens, and were also larger on average. Chapel Hill further maintained a higher price per square foot across bedroom categories, indicating a persistent location-based market premium.
Second, regression analysis of 211 home sales in Ames, Iowa showed that bathrooms, living area, lot size, and home age were significant predictors of sale price, while bedrooms were not significant after controlling for correlated features. The model explained 66.7% of the variation in price, and newer homes were estimated to sell for about $68,706 more than older homes with similar characteristics.
Finally, time-series forecasting of quarterly median home prices in the western United States found that a linear trend model with seasonal effects outperformed a quadratic alternative on validation data and projected continued price growth into late 2025.
Overall, the results show that home value is shaped by both property characteristics and broader market conditions, with location, interior space, bathrooms, home age, and long-term price trends all playing important roles
|
|
More Than a Data Point: Understanding Admissions and Enrollment Conversion
Rose Fotso ( Goldey-Beacom College)*
It is the summer of 2016, and thousands of admitted students stand at the threshold of a life-changing decision: which university acceptance they will ultimately honor. For institutions, however, the central question extends beyond admission itself – of those offered a place, who will actually enroll?
Universities collect large volumes of admissions data, yet understanding what drives admitted students to enroll remains a critical challenge for institutional planning. During the 2016 admissions cycle, 30.7% of applicants received admission offers, but only 8.8% ultimately enrolled, highlighting the importance of post-admission conversion in shaping institutional enrollment outcomes.
This study investigates which academic and demographic factors are associated with the conversion of admitted applicants into enrolled students using a dataset of 17,339 undergraduate applications submitted during the 2016 admissions cycle at a U.S. university. Using descriptive statistics, exploratory visualizations (including histograms and boxplots), and inferential analysis such as two-sample t-tests, the study examines patterns in academic preparedness, admissions outcomes, and enrollment behavior across academic divisions and demographic groups. The applicant pool demonstrates strong academic preparation, with an average high school GPA of 3.56 and a median of 3.59. While gender differences in academic indicators are modest, the analysis indicates that the largest variation in outcomes occurs after admission rather than during the admissions decision itself. Business and Economics programs show the highest enrollment yield, whereas Math and Science programs show lower yield, suggesting stronger external competition for STEM applicants. Differences also appear across demographic and socioeconomic groups, particularly among students from families with lower parental education.
Because key variables such as financial aid, geographic origin, and residency status were unavailable, the findings reflect associations rather than causal relationships. Future research incorporating these variables could support predictive models of enrollment probability, helping universities improve recruitment strategies, enrollment planning, and equitable access to higher education.
|
|
STEM Outreach Across Delaware: Sewing Tomorrow’s Scientists
Oviyanna Umoh (UD)*, Khushdeep Kaur (UD), Taylor Campbell, PhD (UD)
Representation of women in science, technology, engineering, and mathematics (STEM) has improved, with women now comprising 35% of the STEM workforce (National Science Foundation, 2023). Despite this progress, significant underrepresentation persists, particularly among women from historically marginalized communities. Dover High School, a diverse public high school in Delaware, serves a student population that is 55.3% Black, 17% Hispanic, and 43% low-income.
STEM Outreach Across Delaware (SOAD) is an initiative designed to connect underrepresented high school girls with undergraduate mentors at the University of Delaware through a year-long STEM mentorship and science fair program.
During its pilot year, three high school students from Dover High School were paired with undergraduate mentors. One student participated in the New Castle County Science Fair, earning first place in the Behavioral and Social Sciences category. The remaining two students will publish blogs through Project Brain Light, a graduate student–led STEM outreach organization at the University of Delaware, and present their work at Dover High School’s AP Research showcase.
SOAD promotes exploration of STEM fields, builds confidence in scientific skills, and encourages pathways to higher education. This study presents participant impact data from both mentors and mentees, alongside a data-informed growth plan for program expansion. In the coming years, SOAD aims to scale to multiple underrepresented high schools across Delaware, offer regular campus tours, and host an independent science fair at the University of Delaware
|
