Digital Commons @ Florida Atlantic
Not a member yet
400 research outputs found
Sort by
PREDICTIVE AND GENERATIVE MODELING FOR BIOMEDICAL DATA ANALYTICS
Full text linkBiomedical data analytics is a broad interdisciplinary field that has seen vast amounts of data collected with recent technological advancements. Computational studies are required for time and cost-efficient methods to characterize, summarize, and interpret these datasets to advance public health. This thesis presents research studies done with predictive and generative models with regards to subdomains in clinical research, healthcare and bioinformatics. Our research with predictive modeling presents (1) extensive feature engineering methods, ensemble models and feature selection ranking in order to provide data-driven models to evaluate relevant evidence for clinical decision making; and (2) network analysis and unsupervised link prediction methods for clinical trial recommendation. With generative modeling, we present usage of a generative large language model (LLM) to predict novel mutation prediction. Our studies present a variety of methods to improve common classification algorithms on biomedical data and novel methods of representing, classifying and generating biomedical data.
A summary of the contributions of this thesis are as follows:
• We use machine learning methods to investigate prematurely ended clinical trials and address two fundamental questions: (1) what are common factors/markers associated to terminated clinical trials? and (2) how to accurately predict whether a clinical trial may be terminated or not? We introduce extensive feature engineering methods and feature ranking methods to address the first question. For the second question, we train multiple machine learning models with random undersampling and ensemble methods to handle the class imbalance problem leading to satisfactory prediction results which can directly estimate the chance of success of a clinical trial in order to minimize costs. We conduct these studies first with a global (non-disease specific) clinical trial dataset and conduct a COVID-19 clinical trial ablation study with only COVID-19 clinical trials. Results from the global clinical trial study give insights to different research areas as well as general trial attributes that lead to a higher risk of prematurely ending. Results from the COVID-19 ablation study give insights to specific interventions that lead to higher risk of trials prematurely ending.
• We present predictive COVID-19 diagnostic models that predict COVID-19 positive test results from basic symptom and demographic information. We investigate which types of symptoms are highly informative of COVID-19 prediction and show symptom based diagnostic models can predict COVID-19 positive test results. We also investigate the correlation of different COVID-19 diagnostic tests. Outcomes of this study provide information on relationships of COVID-19 diagnostic tests and show that symptom data can be used to predict COVID-19 test results.
• We introduce network analytics of infectious disease clinical trials in order to characterize and understand infectious disease research. We model infectious disease clinical trials as a bipartite network of sponsors and research areas. The network allows us to analyze trends in infectious disease research as well as create a clinical research information recommendation system. Since the network can be sparse, we identify sponsor communities and research area topics. We present a new link prediction method, that utilizes communities and topics, allowing us to recommend related research areas to sponsors
EXPLORING THE RELATIONSHIP BETWEEN HEALTH LITERACY AND HEALTH BEHAVIOR INTENTION THROUGH MEDIATION MODERATION ANALYSES
Full text linkVaccine hesitancy is a complicated public health issue that could compromise the well-being of our communities, evident not only in the COVID-19 pandemic but also in the increasing occurrences of vaccine hesitancy toward long-established vaccines, leading to breakthrough cases of diseases such as measles. Several factors can impact this complex phenomenon, such as health education (HE), attitudes towards social media (SMA) and exercise (EA), chronic disease exposure/risk perception (CDR), and demographic variables.
This study examined the relationship between health literacy (HL), and health behavior intention. The eHeals assessment measured HL and an assessment of vaccine attitude (VA) measured health behavior intention. The researcher delivered an anonymous survey, hosted on Qualtrics, through email to nine health-related courses, which collected responses over 4 weeks. A sample size of n = 909 was accessed through those courses, yielding 592 responses.
The researcher analyzed data through SPSS version 29.2.0 with an add-in macro (PROCESS v4.2 by Andrew F. Hayes, 2022) used to conduct the moderation/mediation analysis. The results indicated a partial mediation of SMA and CDR on the relationship between HL and VA, full mediation occurred with HE, and no mediation occurred with EA. Regarding demographics, the study noted significant moderation in some cases, including number of children, religion, hours worked per week, and COVID-19 vaccine status. The study noted insignificant moderation in age, race, relationship status, and social class variables.
The results demonstrated the complex relationship of these variables and demonstrated that there could be an improvement in VA by strengthening SMA and that past medical experiences and disease risk perception can also increase VA. The role of HE and EA needs to be studied further, as these relationships are not fully understood. Demographic data also need to be examined, as they play a moderating role in some instances
APPLICATIONS OF DEEP LEARNING AND SIGNAL PROCESSING IN ACOUSTICS, OCEANOGRAPHY, AND BIOMETRICS
Full text linkThis dissertation examines multidisciplinary methods in marine research, computational modeling, and artificial intelligence. This dissertation tackles significant issues in ecological monitoring, ocean dynamics forecasting, and assistive technology via three independent but interrelated research streams.
The first study examines the localization and distribution of marine organisms along Florida’s East Coast via passive acoustic monitoring methods. The research analyzed more than 65,000 audio recordings to delineate the geographical distribution of several marine creatures, such as black drum, toadfish, Sei whales, North Atlantic right whales, dolphins, and false killer whales. A novel automated detection and localization model was developed, using an adaptive matching filter and Time Difference of Arrival (TDOA) algorithm, which demonstrated accurate marine species tracking with localization errors of about 2 meters across distances of 50 meters.
The second research enhances oceanic modeling with a hybrid deep learning methodology that integrates Empirical Orthogonal Function (EOF) analysis with a Fourier Neural Operator (FNO). This novel technique significantly improves the prediction of ocean velocity fields, demonstrating exceptional performance across several datasets. The suggested EOF-FNO model has exceptional zero-shot super-resolution capabilities and shows improved stability in capturing intricate spatiotemporal marine dynamics, offering a viable computational approach for comprehending oceanic systems.
The last research examines deep learning applications in biometrics, with a particular focus on communication accessibility and security technologies. The research attained exceptional performance in American Sign Language (ASL) recognition and Finger Knuckle Print (FKP) classification via extensive comparative analyses of several neural network architectures. The Vision Mamba (ViM) models have shown exceptional performance, with an accuracy of 99.98% in ASL recognition and 99.1% in biometric identification, underscoring the capabilities of modern deep learning methodologies in assistive and security applications.
These research efforts bring advanced computational approaches that integrate ecological monitoring, deep learning, and biometric innovation. This dissertation integrates modern sensing technologies, complex algorithms, and deep learning methods to provide unique insights and tools for comprehending marine ecosystems, forecasting oceanic incidents, and creating dependable assistive technologies. This work’s highlights the capacity of the advanced computational methods in tackling intricate scientific and technical issues
USER BEHAVIOR ANOMALY DETECTION APPROACHES TO MITIGATE INSIDER THREATS
Full text linkInsider threats represent a critical challenge in cybersecurity due to their ability to disguise malicious activity within legitimate user behavior. This dissertation proposes novel approaches to enhance insider threat mitigation through user behavior anomaly detection. First, a systematic survey is conducted to evaluate existing countermeasures, categorizing technical and human-centric strategies while identifying their limitations. building upon these findings, the dissertation introduces two complementary detection frameworks: the contiguous, contextual, and classifying pipeline (C3P), which uses symbolic pattern mining and contextual modeling to autonomously score and classify sequences, and the representation-reconstruction detection (R2D) framework, which leverages causal self-attention and variational autoencoding to identify anomalies in latent space. together, these approaches address key challenges of scalability, limited contextual understanding, and data labeling dependency, providing a more adaptive, interpretable, and robust solution for detecting insider threats in complex user action sequences
PSYCHOMETRIC PROPERTIES OF A NEW ACCULTURATION SCALE FOR LATINOS/HISPANICS
Full text linkThe current study evaluated the psychometric properties of the newly developed Multi-Ethnic Acculturation Scale (MAS), a multidimensional instrument designed to assess behavioral and cognitive aspects of cultural adaptation among Hispanic/Latino individuals in the United States. A sample of 208 adults from diverse Hispanic backgrounds completed the MAS alongside two validated acculturation measures: the Short Acculturation Scale for Hispanics and the Vancouver Index of Acculturation. Exploratory factor analyses revealed a partially distinct but moderately correlated factor structure for both American and Latino cultural orientation subscales, supporting a bidimensional conceptualization of acculturation. Internal consistency was high across both subscales (Cronbach’s α \u3e .90), and test-retest reliability demonstrated temporal stability. Regression analyses showed that proficiency and language dominance significantly predicted overall acculturation scores. Overall, these findings support the MAS as a reliable and valid tool for assessing acculturation across diverse Hispanic populations
ANALYSIS OF THE 2-PERSON COMBINATORIAL GAMES SPLIT-S-NIM AND CHOMP ON 2 ROWS
Full text linkSplit-S-Nim is a variant of Nim in which each player on their turn can choose to make any legal move in the traditional version of Nim or add q coins to a heap, where q is some element of a predetermined subset of integers, S. We classify all sets that have the property in which the winning strategy is equivalent to the winning strategy for a version of the game with a set that has cardinality 1.
If S has a smallest non-negative even value, q, then we conclude that it must either have a winning strategy identical to a version of the game with the set S = {q} or must have a general winning strategy that differs from any general winning strategy appropriate for a version of the game such that S has one element. This winning strategy is found by calculating the Sprague-Grundy numbers for the game with one heap.
If S has no non-negative even elements, we show that the winning strategy is the same as the traditional game of Nim. Similarly, we show that if there is an odd integer in S greater than or equal to −1, then the winning strategy must differ, with the exception of the odd value of 1 when 0 is in S.
Finally, we show subsets with smallest non-negative even elements of the form 2n + 4s or 2n + 4s + 2 will only have a winning strategy identical to a singleton set version of the game if all other even values have certain properties.
We complete our study of Split-S-Nim by considering the winning strategy for a game where S = Z. This version of the game is a logical extension of the game Nim in which the player can replace a heap of any size with up to two heaps of a size smaller than the original heap.
Chomp is a combinatorial game attributed to Frederik Shue and David Gale in which players take turns removing rectangular pieces from an n × m grid. While a winning strategy for the first player has been shown to exist, the strategy is not known. We analyze the case where the starting position has 2 rows, finding the Sprague-Grundy numbers for all subgames of that starting position
CROWDING OUT WHISTLEBLOWER ALTRUISM: EXAMINING THE IMPACT OF WHISTLEBLOWER INCENTIVES AND RECOGNITION ON RETALIATORY BEHAVIORS
Full text linkWhistleblowers play an instrumental role in exposing organizational wrongdoing and, thereby, helping to remedy the damage endured by stakeholders. However, whistleblower retaliation is rising, its personal impacts to the whistleblower are severe, and fears of retaliation serve as a strong deterrent to prospective whistleblowers. In this dissertation, I draw on motivation crowding theory, which posits that external interventions can crowd out or crowd in intrinsic motivations for engaging in a behavior. I employ a third-party perspective of motivation crowding theory and utilize a 2x3x2 between-participants experimental design to examine factors that impact employees’ perceptions of whistleblower intent and acts of retaliation. Results indicate when the reported fraud would have benefited a group versus only the perpetrator, the whistleblower is viewed as less altruistic, and both managerial and coworker retaliation are more likely. Therefore, it is particularly important to identify tactics that are effective in improving employees’ altruistic whistleblower perceptions and mitigating whistleblower retaliation among these group-benefiting frauds. I find, within the group-benefiting reported fraud, whistleblower recognition is an effective means of improving employees’ whistleblower perceptions when the whistleblower received either a cash reward or no reward for reporting. While whistleblower recognition for group-benefiting reported frauds is primarily effective in reducing managerial retaliation when the whistleblower received either a tangible reward or no reward, it is most effective at reducing coworker retaliation when the whistleblower was rewarded with cash. Within individual-benefiting reported frauds, my findings are more limited but offer support for the beneficial impact whistleblower recognition can have on reducing managerial retaliation when the whistleblower receives no reward or a cash reward, and coworker retaliation when the whistleblower receives a cash reward. My retaliation findings suggest that the drivers of managerial and coworker retaliation are different, thus, it is valuable to consider the impact company programs and responses have on both levels of retaliation
MODELING AND NUMERICAL SIMULATION OF A LIFTING SURFACE CONTROLLED OCEAN CURRENT TURBINE
Full text linkAs a relatively novel source of energy extraction, many design concepts for ocean current turbines (OCTs) still lack the analyses required to mature their technological readiness. This study addresses this issue as it provides a numerical simulation tool for lifting surface controlled OCTs. The performance of a 1.4 MW moored, dual-rotor, lifting surface-controlled OCT operating in a total water depth of 375 m is then numerically evaluated using this tool. Algorithms are based on published OCT research with modifications developed for the present design. Performance is evaluated at varied flow velocities, turbulence intensities, and control surface positions to represent normal and extreme conditions in the ocean with open-loop control. A closed-loop PID controller is tuned and used to reevaluate performance. The performance is characterized by the resulting operating ranges, Euler angles, and power generation. This work provides a basis for future optimization and development of the lifting surface-controlled OCT desig
TOP MANAGEMENT TEAM COMPOSITION AND FINANCIAL REPORTING QUALITY: A FAULTLINE PERSPECTIVE
Full text linkThis study examines whether faultlines within top management teams (TMT) influence financial reporting quality (FRQ). Faultlines reflect hypothetical divisions within a group based on team members’ unique demographic characteristics and experiences. Using a large sample of U.S. firms from 2003 to 2020, I construct a faultline measure based on executive gender, age, tenure, education, and board experience. I test whether faultlines negatively associate with FRQ, proxied by measures of restatements and accruals quality. I further investigate faultline type, based on the demographic (gender and age) and task-related (tenure, education, and board experience) components. The main results do not support the prediction that faultlines, whether overall, demographic, or task-related, affect FRQ. However, in cross-sectional settings, I find some evidence that demographic faultlines associate with improved FRQ when complexity is low. This contrasts with previous research that suggests faultlines are detrimental to team outcomes. This paper contributes to the mixed evidence on TMT characteristics, and I call for future research to adopt more standardized measures of the TMT and encourage replications