ScholarWorks (California State University)
Not a member yet
87523 research outputs found
Sort by
Automated Document Classification and Data Extraction System Using Machine Learning: A Case Study for Academic Document Management
No full textThis research presents the design, implementation, and evaluation of a machine learning-based system for automating the classification and data extraction of academic documents, specifically high school and college transcripts. The system integrates traditional classification models, including Logistic Regression, Random Forest, and Support Vector Machines (SVM), with advanced Natural Language Processing (NLP) techniques for named entity recognition (NER), using tools such as spaCy and transformer-based models like BERT, RoBERTa, and DeBERTa. Unlike previous studies that rely on synthetic or publicly available datasets, this research uses real academic transcripts collected from a university setting. Results show that while all classification models achieved perfect accuracy, SVM and Random Forest consistently demonstrated higher confidence in their predictions. For data extraction, transformer models performed well on transcripts with layouts seen during training, with accuracy declining on transcripts that had unfamiliar layouts. Overall, the system reduces manual workload, improves accuracy, and offers a scalable solution for academic institutions seeking to automate document processing
Proteomic Determinants of Embryonic Resistance to Heat Shock in Petrolisthes cinctipes
No full textThis study presents the first comprehensive proteomic analysis of embryonic stages in a decapod crustacean, the porcelain crab (Petrolisthes cinctipes), providing new insights into how environmental variability shapes developmental biology. Proteomic profiles from embryos reared in dynamic field conditions at Fort Ross, California, were compared to those from embryos cultured under laboratory conditions. Field-collected embryos exhibited distinct proteomic adaptations, characterized by elevated expression of proteins involved in energy metabolism, immune responses, oxidative stress defense, and structural integrity. Notable proteins enriched under natural conditions included acetyl-CoA acetyltransferase, alpha-2 macroglobulin, diverse vitellogenin isoforms, and salt-inducible kinases. These proteins appear integral to the embryos' ability to cope with intertidal stresses such as temperature and salinity fluctuations. Functional analyses using KEGG pathways and STRING interaction networks confirmed the biological coherence of these adaptive proteomic adjustments. Furthermore, differential expression of vitellogenin isoforms indicated sophisticated maternal provisioning strategies tailored to environmental stressors. Methodologically, this research established a robust shotgun proteomics workflow and generated a novel spectral library for P. cinctipes, significantly enhancing future ecological and physiological studies of non-model organisms. These findings provide a critical foundation for understanding how intertidal populations may respond to climate-driven environmental changes.https://doi.org/10.46569/ws859r88
College Enrollment Decisions and Trajectories Among Community College Applicants
No full textUnderstanding how California Community College applicants decide to enroll is essential for increasing admission yield rates at these colleges. This explanatory, sequential mixed-methods study examined how applicants' knowledge, perceptions, and resources shaped their decisions about college attendance. The first phase included a quantitative analysis of the higher education enrollment outcomes of 35,100 applicants who applied to a California Community College between 2019–2020 and 2023–2024, including enrollment in a community college, a 4-year college or university, or no higher education institution. Additionally, chi-square tests of independence and analysis of variance were conducted on data from 6,671 applicants from the 2023–2024 academic year, revealing significant associations between applicants' demographic, educational, and socioeconomic characteristics and their enrollment outcomes. Findings from this phase informed the second, qualitative phase, which involved interviews with 13 individuals to explore how they made their decisions about college enrollment. Results indicated that families, friends/peers, college reputation, high school teachers, transfer opportunities to a university, the COVID-19 pandemic, educational intent, cost/affordability, college location, work, and the class registration process all played a pivotal role in shaping decisions. However, decision making was constrained by bounded rationality (Simon, 1955), as applicants relied on incomplete information, heuristics, and perceived accessibility rather than fully rational comparisons. Additionally, findings aligned with Iloh's (2018) model of college-going decisions and trajectories, demonstrating how students navigated enrollment pathways based on structural constraints, lived experiences, and access to social and institutional capital. This study contributes to the literature by highlighting the interplay between cognitive limitations, structural barriers, and individual agency in enrollment decision making. Findings have implications for policy and practice.https://doi.org/10.46569/8k71ns75
The Price of Protection: How School Safety Measures Reinforce Racial and Spatial Inequalities
No full textThis quantitative, correlational study employed critical race spatial analysis (CRSA) to investigate how actual neighborhood crime rates and families' subjective perceptions of neighborhood violence—shaped by historical patterns of racial and spatial inequality—influence confidence in standardized school safety procedures and the emotional well-being of families in urban elementary school communities. Survey data were collected from parents and caregivers, capturing demographic details (e.g., race/ethnicity, income, education level), neighborhood tenure, and perceptions of local violence. These responses were then linked with official crime statistics to illuminate the relationships among neighborhood conditions, institutional trust, and families' evaluations of lockdown LD, a standardized procedure in which the school is secured in response to an immediate threat, and shelter-in-place measures, a strategy to keep students and staff inside during a potentially hazardous external event. Findings indicated that perceived violence exerts a stronger influence on families' confidence than do official crime metrics—a pattern especially pronounced among nonwhite families in historically under-resourced neighborhoods. Yet, these racialized dynamics are complex; at extremely high levels of perceived violence, White families' emotional distress rises to match or exceed that of some racially marginalized groups, suggesting that perceptions of local threats can reshape usual patterns of privilege. Although trust in school administration has consistently predicted higher confidence in safety protocols, its hypothesized moderating effect—where trust might buffer the emotional toll of neighborhood conditions—has remained less robust. Instead, factors such as race, education, and drill frequency more reliablyvexplain why certain families experience significant anxiety while others remain resilient. These results suggest one-size-fits-all safety protocols may be insufficient in communities shaped by decades of neglect and structural racism. Instead, context-sensitive interventions that prioritize relationship-building, transparent communication, and collaborative policymaking can foster genuine trust and mitigate emotional distress. By centering family perspectives and acknowledging the pivotal role of race and space, this dissertation contributes to a more nuanced understanding of school safety—one that moves beyond official crime rates to address the lived realities of those most affected by neighborhood conditions.https://doi.org/10.46569/ht24wv36
Hierarchy or Education: Why is Trauma Informed Pedagogy Vital to the Educational Program's Outcomes for Black Males in Post-secondary Education?
No full textThis thesis will investigate the hegemonic ideological framework that defines the continuous development of colonial oppression of Black males through education and the trauma they are subjected to. In doing so, it investigates the effectiveness of postsecondary programs that are designed to aid and assist Black students as well as examines the historical trauma that Black people, as a community, have been exposed to in order to understand the social experience of Black students prior to postsecondary education. As Black is a racial construct (Coates, Brunsma, & Ferber, 2018), for the purpose of this research I will use that term to define the people of color who have been colonized and are behind enemy lines politically and physically. Those who dwell behind enemy lines are defined as those who are oppressed politically, economically, physically, emotionally and spiritually by structured racism and violence (Cabral, 1973; Noel, 1968; Ritzer & Stepnisky, 2018; ). The U.S. educational system was not designed to nurture and support Black students. On the contrary, its nocuous policies were designed by centuries of colonization and imperialism. The methods I will use will be a mixture of quantitative surveys (questionnaire) of Black students and qualitative (interviews) of faculty (Du Bois, 2007). The American Black Nationalist Epistemological Methodology is my ideological framework for this project. The aim of the research is to identify overarching themes and challenges that could be barriers to the program's effectiveness and success of Black male students in order to develop strategies to counter the agitation and exacerbated conditions of learning in a colonial system. Because the residual trauma cannot be ignored, trauma informed pedagogy is necessary to ameliorate emotional hidden suffering. This research will benefit individuals, university programs and communities. It can bring a heightened awareness of the social conditions that historically have deleterious impacts on minority communities, in particular the Black community.https://doi.org/10.46569/ns064h16
Comparing recidivism rates of female offenders in specialized reentry programs and traditional parole supervision: A systematic literature review
No full textStudy Purpose: This study assessed gender-responsive reentry programs versus traditional parole in reducing women's recidivism by analyzing evidence on housing, employment, substance abuse treatment, and mental health services. Methods: The study review analyzed 20 diverse methodological studies across five databases using strategic search terms related to female correctional programming. Findings: Female inmates show better outcomes with gender-responsive approaches addressing mental health, substance abuse, and trauma, while vocational training and specialized interventions help mothers overcome employment barriers and reintegration challenges. Discussion: Gender-neutral correctional policies fail to address women's unique needs. Effective recidivism reduction requires comprehensive trauma-informed approaches integrating mental health services, substance abuse treatment, vocational training, and family reunification support
Learning-based matheuristic approaches to the University Course Timetabling Problem
No full textThe University Course Timetabling Problem (UCTP) concerns the assignment of instructors, classrooms, and timeslots to academic courses, subject to a complex system of constraints. Like other scheduling problems prevalent in transportation, manufacturing, healthcare, etc., the UCTP is combinatorially complex and is generally classified as NP-hard; certain formulations are known to be NP-Complete (Yang et al., 2017). While the UCTP can be modeled as a constrained feasibility problem via integer programming methods, the introduction of a secondary scoring mechanism via an objective function allows us to pose the UCTP as an optimization problem -- enabling the search for high-quality solutions within our search space. This thesis presents a Matheuristic framework for the UCTP -- we create a hybrid optimization strategy which we tailor to our problem, integrating integer linear programming with selected metaheuristics in stages to find heuristic solutions to our problem, consistent with the definition presented by Maniezzo et al. (2021). In particular, we use Genetic Algorithms and Simulated Annealing to construct a learning-based search process with solution repair capabilities. This iterative refinement mechanism, informed and inspired by evolutionary processes at both individual and generational levels, allows us to navigate our ILP decision system intelligently -- specifically, a we design and tailor a heuristic learning algorithm which improves solution quality iteratively. We demonstrate that our methodology efficiently finds multiple stable, near-optimal schedules, and was able to reduce the effective search space of a sample problem from o(10^151) to below o(10^6). These results showcase the potency of hybridizing mathematical programming with metaheuristics in solving large-scale, real-world combinatorial optimization problems
David McCullough's Narrative
No full textDavid Mccullough is considered one of the best historical writers when it comes to writing American history because he not only brings history to life but also tends to be more descriptive in his work with three elements; character, setting, and theme. Mccullough has this great sense of knowledge about history but also his ability to use narration to write about history in the non- fiction sense. Which is what I will be discussing in my research paper as to how David Mccullough uses fictional narration to write non-fictional narration when it comes to writing American history. He is always describing a setting or a character in most of his books and makes sure that the theme of the story is there. Mccullough wants the reader to be interested in his books than just reading them for information
GA-SFS: A two-stage hybrid algorithm for feature selection in biomedical data
No full textIn biomedical data analysis, feature selection is crucial, particularly for high-dimensional datasets where redundant or irrelevant features might affect model performance. Biomedical datasets introduce additional challenges, such as high dimensionality, small sample sizes, and complex data structures which will then lead to overfitting, increased computation costs, and having trouble generalizing results. Addressing these issues often requires hybrid, adaptive, or ensemble-based algorithms that can improve model robustness and interpretability. To meet these demands, this paper introduces GA-SFS, a hybrid algorithm that combines Genetic Algorithms (GA) with Sequential Forward Selection (SFS) to select informative features in complex biomedical datasets effectively. GA explores the global search space to recognize relevant feature sets, while SFS refines the subsets by iteratively adding and removing features to have a more accurate classification performance. Experimental results on multiple datasets including real world biomedical applications, show that GA-SFS outperforms traditional feature selection methods in classification accuracy while reducing redundant or irrelevant features. These findings highlight the potential of GA-SFS as a reliable and effective feature selection approach in biomedical research. Index Terms-Feature Selection, Genetic Algorithm, Sequential Forward Selection, Biomedical Datasets, Hybrid Mode
Smart Hydration: A Smartwatch-Based System for Water Intake Monitoring and Drinking Gesture Recognition
No full textThis study investigates using smartwatches to track water intake by recognizing drinking gestures and estimating fluid volume. The problem addressed is the challenge of automating hydration tracking in daily life, using readily available wearable technology. The system utilizes accelerometer and gyroscope data, combined with Dynamic Time Warping, to distinguish drinking motions from other common arm movements like chopping or waving. The method achieved 100% accuracy in identifying drinking, chopping, and waving gestures, but struggled with punching (76.7%) and resting (33.3%), likely due to overlapping motion patterns. For volume estimation, various machine learning models were tested, with Long Short-Term Memory and Gated Recurrent Units achieving the best performance, resulting in a Mean Absolute Error of approximately 4.60 milliliters. These results demonstrate the feasibility of using smartwatch sensors for both gesture recognition and fluid volume estimation, highlighting the potential of machine learning in improving hydration tracking