ScholarWorks (California State University)
Not a member yet
87523 research outputs found
Sort by
Bagging or Boosting? Fraud Detection in Financial Transactions
No full textThis study systematically evaluates Bagging and Boosting techniques - Random Forest, Extra Tree, Isolation Forest, and AdaBoost, Gradient Boost, XGBoost, respectively - for fraud detection. After identifying the best-performing method from each type, the study further applies local outlier factor to their results to improve anomaly identification and refine classification accuracy. The models are trained and tested on real-world financial transaction datasets, with their performance assessed using performance metrics that include precision, recall, and F1-score. The purpose of this study is to compare these ensemble approaches in order to shed light on their benefits and limitations for fraud detection. The study explores how the various techniques perform upon datasets with varying characteristics and operational constraints, providing assistance for choosing the most suitable approach for real-world financial security applications. The findings of this study contribute to the broader understanding of ensemble learning in fraud detection and its practical implications in the financial sector
Enhancing diabetes detection using machine learning
No full textDiabetes is a chronic condition that impairs the body's ability to regulate blood sugar levels, leading to serious health complications if left untreated. With its rising prevalence worldwide, early detection of diabetes is critical for effective management and prevention of long-term damage. This thesis aims to enhance diabetes detection by applying advanced machine learning techniques, which can improve predictive accuracy and assist in early diagnosis. The Pima Indians Diabetes Database is used to develop and compare various models, including Logistic Regression, Random Forest, Support Vector Machine (SVM) with a linear kernel, K-Nearest Neighbors (KNN), Decision Tree, Gradient Boosting, Naive Bayes, XGBoost, MLP Neural Network, AdaBoost, LightGBM, CatBoost, and Ridge Classifier. By evaluating these models through metrics such as accuracy, precision, recall, F1 score, and AUC-ROC, this research seeks to identify the most effective algorithms for diabetes prediction. The study not only demonstrates the significance of machine learning in improving detection rates but also highlights its potential to reduce healthcare costs and enable personalized treatment plans. This research can lead to more timely interventions and better patient outcomes, offering a valuable contribution to the field of diabetes management and overall healthcare improvements
Machine Learning Approaches to the Analysis of Extremophile DNA
No full textMachine learning makes it possible to identify associations between different features in data that would be challenging — or even impossible — for a human investigator to find, particularly in very large, complex data sets such as genetic data. This thesis uses decision-tree-based machine learning techniques to analyze metagenomic data (specifically the frequencies of taxonomic classes) in order to predict features of the environment from which the genetic samples originated — a novel approach. We evaluate the efficacy and efficiency of three particular decision-tree-based approaches to this analysis — Random Forest, XGBoost, and CatBoost — to explore these relationships in the context of samples containing genetic data from so-called extremophile life (we focus on organisms that thrive in extreme heat, cold, and pH). Encouragingly, all three decision-tree-based approaches can be used to successfully predict the pH and temperature of the environment based on frequencies of taxonomic classes represented in the metagenome of a given sample. Overall, CatBoost outperformed Random Forest and XGBoost in predictive success, but not temporal efficiency. This thesis employs best practices for ethical and explainable AI to audit the data and to characterize the results.https://doi.org/10.46569/6682xd99
Developing Differentiated and Sustainable Professional Learning Communities Among Chinese Immersion Preschool Teachers
No full textThe current dissertation investigated how professional learning communities (PLCs) can be developed and sustained to support Chinese immersion preschool teachers across diverse roles, including lead teachers (LTs), teaching associates (TAs), and support teachers (STs), in an independent school context. Through the lens of educational equity and teacher empowerment, the study explored how differentiated PLC models can respond to the professional development (PD) needs of immigrant educators while enhancing individualized, child-centered learning practices. Using design-based implementation research (DBIR) methodology and narrative inquiry, the current study drew on stories and experiences spanning from 2020 to February 2025, and formal data collection began in April 2024. Grounded in theoretical frameworks such as communities of practice, complexity leadership theory (CLT), design thinking, multitiered system of supports (MTSS) and universal design for learning (UDL), the research examined three iterative cycles of PLC implementation. The findings illustrated how leadership practices, collaborative structures, and reflective documentation (e.g., learning stories) can foster teacher growth and systemic change, particularly in multilingual, multicultural early learning environments. The current study contributes to the field by proposing a sustainable, inclusive PLC model that bridges hierarchical gaps in teacher roles, integrates culturally responsive pedagogy, and enhances support for all children's developmental needs. The results offer actionable insights for educational leaders and policymakers seeking to improve equity, collaboration, and instructional quality in early childhood education (ECE), especially in bilingual and immersion program.https://doi.org/10.46569/mp48sp50
The Act of Telling, the Work of Becoming: Multiracial Teacher Identity Development
No full textThis qualitative dissertation examined how multiracial teachers constructed and articulated their racial and professional identities through storytelling and narrative. As the multiracial population in the United States grows, including the number of multiracial educators, schools continue to operate under monoracial assumptions that often render multiracial identities invisible or misunderstood. Drawing on autoethnography and narrative inquiry, the study centered on the lived experiences of 20 multiracial educators, including the researcher's own experience, to examine how multiraciality shaped teacher identity, pedagogy, and interactions with school communities. The research was guided by three questions: (a) What narratives do multiracial teachers construct about their racial identities? (b) How are these narratives articulated in relation to their teaching? (c) How do these teachers perceive and shape narratives of racial identity in their schools? Findings highlighted the fluid and context-dependent nature of multiracial identity, tension between personal and ascribed racial categories, and emotional labor of navigating institutional structures that promote monoracial norms. The study proposed the need for educational institutions to adopt inclusive practices that validate multiracial experiences, invest in identity-affirming professional development, and rethink how teacher identity is supported across racial lines. This research contributed to emerging work in multiracial studies and offered implications for self-reflection, teacher preparation, curriculum, and professional development.https://doi.org/10.46569/np193k98
Meet Rack
No full textSet in Tucson aka the Dirty T, Meet Rack is about a guy who falls for the wrong girl. Based on the playwright's life during the ugly side of their twenties, but a work of fiction.https://doi.org/10.46569/bz60d614
Exploring the Collaboration Challenges Faced by AI Researchers and Health Experts for Developing Computational Techniques for Healthcare Systems
No full textArtificial intelligence (AI) and machine learning (ML) are reshaping the landscape of healthcare, offering profound opportunities to augment clinical decision-making, streamline operations, and enhance patient outcomes. From diagnostic imaging to predictive analytics and personalized treatment planning, AI tools promise transformative change. However, despite significant technical advances, the meaningful integration of AI into clinical practice remains slow and uneven. This gap between potential and real-world application is not solely due to technological limitations, but often stems from deeper human, organizational, and collaborative barriers. This thesis investigates the collaboration processes and challenges faced by AI researchers and healthcare domain experts in developing computational techniques for healthcare systems. Through a qualitative research design, I conducted semi-structured interviews with fifteen participants, ten AI researchers and five healthcare professionals—who were actively engaged in interdisciplinary healthcare AI projects. Participants were selected through purposive sampling to ensure firsthand experience at the intersection of AI development and clinical practice. Thematic analysis, supported by the Lightning Synthesis method, was employed to extract and synthesize key themes. The findings were structured using a four-stage collaboration frameworkencompassing Defining the Problem, Data Collection and Preprocessing, Model Training and Quality Control, and Evaluation. The study reveals that collaboration barriers are deeply rooted in cognitive mismatches, workflow incompatibilities, domain knowledge translation challenges, and misaligned success metrics. AI researchers often struggled to translate vague or context-specific clinical insights into structured data inputs, while clinicians faced difficulty understanding technical constraints and assumptions embedded in AI model development. Differences in reasoning styles—quantitative versus qualitative, granular versus holistic—further complicated mutual understanding. Trust, an essential ingredient for interdisciplinary partnership, was often fragile and had to be consciously cultivated through transparency, iterative communication, and shared evaluation frameworks. Clinician perspectives highlighted additional challenges, including time constraints, skepticism toward AI reliability, concerns about workflow integration, and fears regarding loss of professional agency. Nevertheless, both groups recognized the potential benefits of closer collaboration. Participants identified promising strategies such as embedded observational learning, co-design of data representations, use of visual explanation tools, structured feedback forms, and broader definitions of success that encompass both technical accuracy and clinical utility.This thesis contributes to the growing field of human-centered AI in healthcare by offering a dual-perspective, process-level understanding of interdisciplinary collaboration. It builds upon prior theoretical frameworks of hybrid intelligence and third spaces by grounding them in empirical evidence from real-world AI-healthcare projects. Practical recommendations are offered for AI research teams, clinicians, tool developers, and institutions seeking to foster more resilient, inclusive, and impactful collaborations. Ultimately, this work underscores that the future success of AI in healthcare will depend as much on human relationships, communication, and trust as on algorithmic innovation.https://doi.org/10.46569/h702qg62
Identifying H-α Dimming in Solar Pre-Flare Events
No full textSolar flares are a magnificent display of the Sun's raw power. Flares and their asso- ciated coronal mass ejections send enormous amounts of ionized particles and harmful radiation into the solar system at relativist speeds. As the high-speed material collides with Earth's magnetosphere, they can create breathtaking aurora displays in our polar regions. However, they can also induce geomagnetic storms, disrupting satellite communications, GPS navigation, and power grids. Solar flares pose a serious threat to modern technologi- cal systems therefore, to better understand the details of flare physics this research aims to identify pre-flare dimming in solar flare light curves. In future work, this information could be used as a potential indicator for an imminent flare. I obtained Rapid Oscillation in the Solar Atmosphere (ROSA) data of a M3.9-Class solar flare event that occurred on June 11, 2014, at 19:19:21 UTC and analyzed the pre-flare behavior. I also obtained two additional flares, an M1.1 from June 12, 2012 and C2.3 from July 8, 2011. I created light curves to measure the flare properties such as peak emission rise and decay times in an attempt to identify solar pre-flare dimming in the light curves. Comparison to flare light curve simulations showed the Balmer line dimming over 10's of seconds and an impulsive absorption for fractions of a second. However, the M3.9-Class solar flare showed a dimming period of hundreds of seconds before the solar flare entered the impulsive phase and reached its peak. Then, I confirmed this occurrence with the peak- normalized time evolution of continuum for a September 2014 light curve of a HF/GF-type dMe solar flare. The two solar flare events demonstrated dimming during the pre-flare phase and for a longer duration of time than expected. While the current data suggests that pre-flare dimming may serve as a good test of solar flare physics, in future work I can study if this can be used as a reliable indicator of an impending solar flare. Further comprehensive high time resolution observations, analysis and modeling are essential to confidently establish this analytical and predictive value for flare events
Trends and Risk Factors Among a Decade of Occupational Heat Fatalities
No full textSince 2014, the National Council for Occupational Safety and Health (COSH) has published a database of employees who were killed at work, and in 2020, COSH began highlighting the subset of those victims that died from heat stress. In 2021, the Occupational Safety and Health Agency (OSHA) announced that it would consider initial rule-making for a Heat Illness Prevention Standard (HIPS). In the absence of a specific standard, workplace heat fatalities are enforced under the General Duty Clause (GDC). A specific HIPS has been in place in California since 2005, under their approved State Plan, the California Department of Occupational Safety and Health. Fatalities attributed to heat were collected from the Bureau of Labor Statistics (BLS) Census of Fatal Occupational Injuries (CFOI), the COSH US Worker Fatality Database, The Center for Construction Research and Training Construction (CPWR) FACE Database, and the United Support & Memorial for Workplace Fatalities (USMWF) databases. Cases were evaluated for relevance; fatalities that were determined to be due to injury, trauma, or other non-heat causes were eliminated. The remaining cases (N = 249) were used to construct a dataset comprised of CFOI variables (including incident abstracts) and enforcement data obtained from the OSHA Integrated Management Information System (IMIS). Text data were coded for risk factors, preventative measures, and worker vulnerabilities. SPSS was used to conduct statistical analyses on numeric data. Across this decade, cases increased in quantity and geographic spread. Victims of heat stress were found to be mostly male, Hispanic, and laborers. Risk factors include being a non-union worker, working outdoors, and performing a low-wage job. Notably, mentions of risk factors, worker vulnerability, or preventative measures taken were most often lacking in the case abstract, resulting in knowledge gaps in lessons learned, and suggesting an inconsistency among inspectors. Cases cited under California's HIPS resulted in significantly higher fines than those cited under the GDC (for initial fines, p = 0.001; for final fines, p < 0.0001). Heat exposures are likely to increase in severity for outdoor workers as climate change progresses. Priorities for a special emphasis program include Spanish language translation, and efforts to engage the agricultural, construction, and landscaping sectors in prevention programs. Opportunities exist to improve symptom recognition, recognize individual risk factors, implement a buddy system, and provide more effective emergency medical response. More specific standards, such as those detailed in the California HIPS, may contribute to more rigorous enforcement and help investigators apply consistent standards to identify gaps in employer heat stress prevention measures
Scoping Literature Review: Exclusion of Persons with Disabilities
No full textThe purpose of this scoping review is to explore the exclusionary experiences of people with invisible disabilities (ID). While there are many studies highlighting the exclusionary experiences of people with invisible disabilities, no known scoping literature review has been done. This study addresses this gap by identifying key themes in the literature, shortcomings, and areas of future research. The study also evaluates the literature on ADEI. In this scoping literature review, fifteen articles were analyzed. Key findings include exclusion taking place in all places such as school, work, and personal life. Additionally, exclusion is exhibited in different forms such as lack of accommodation, judgement, lack of respect, and perpetuation of ableism. This study aims to further inform policy, social work practice, the collective awareness, and actions towards the experiences and implications of persons with invisible disabilities