22505 research outputs found
Sort by
Beyond the Badge: Predicting Military Employee Attrition in the Police Force
Get PDFThis study explores the issue of employee attrition within the military police force, with a focus on predicting which employees are at risk of leaving. Attrition in law enforcement is costly and disruptive, especially in roles that require long training periods and operational readiness. The research specifically targets the internal organizational factors influencing turnover in military police environments and aims to leverage machine learning models to support the early identification of high-risk individuals. Guided by the CRISP-DM framework, the study followed a structured process across six stages: business understanding, data exploration, preparation, modeling, evaluation, and insight generation. A simulated dataset of 2,500 military employee records was analyzed, including variables such as age, rank, years of service, job satisfaction, fitness readiness, and promotion history. Statistical tests like Chi-Square and Mann-Whitney U were used to identify significant features contributing to attrition. Five machine learning models were built—Logistic Regression, Random Forest, Support Vector Machine, Artificial Neural Network, and XGBoost. Among these, XGBoost outperformed others in recall, F1-score, and accuracy, making it the best model for predicting attrition. Key findings showed that salary, years of service, and educational level were the strongest predictors of attrition, while promotion and fitness readiness played a smaller role than expected. The study concludes that machine learning can be a powerful tool for supporting HR decision-making in law enforcement settings. The models developed in this research can help organizations implement early interventions, tailor retention strategies, and better allocate resources. However, the study was limited to internal data and did not include external factors like economic trends or personal motivations. Future research could expand on this by including broader variables and real institutional data from other military entities across the GCC region
The Impacts of Water Lily Invasion and Removal on Wetland Ecosystem Function
Get PDFColonization by native and non-native invasive plants is considered a primary reason for the failure of wetland restoration and creation projects. We investigated the impact of invasion and subsequent removal of Nymphaea odorata (white water lily) in two permanently flooded wetlands in Western New York State. Long-term grazer exclusion experiments at these sites demonstrated that selective grazing by herbivores, such as waterfowl, reduced emergent vegetation and overall plant diversity, simultaneously facilitating N. odorata expansion. This interaction ultimately promoted a negative feedback to waterfowl use of the wetlands because of the lack of open water space. To evaluate potential remediation options, we experimentally removed N. odorata in both small and large-scale plots and assessed impacts on methane emission, plant diversity, soil characteristics, potential denitrification, and waterbird use. In small plots, N. odorata removal was crossed with grazer exclusion to evaluate interactive effects. In smaller plots, removal resulted in a marginal decrease in N. odorata cover, but only where grazers were excluded. There were no persistent effects among years. However, plant diversity increased in grazed plots with N. odorata removal, trending towards diversity measured in exclusion plots. Soil characteristics, methane flux, and potential denitrification were not impacted by removal efforts. In large zones, bird use increased significantly with removal in spite of the lack of reduction in N. odorata cover. These results highlight the importance of considering multiple drivers of ecosystem functions, including invasive plants and herbivory, during efforts to improve wetland restoration outcomes
Using Predictive Analytics to Reduce Small Business Cost Estimation Error
Get PDFSmall and medium packaging companies generally employ the use of custom-developed quoting programs to bid goods and services. Custom bid programs (e.g. Excel) are used to capture the company-specific costs of production. The inputs of variable costs, such as machine rate and scrap rate, are critical to get correct; however, companies often rely on educated guesses and industry expertise to quote packaging products to end-users. Due to the guesswork involved there can be a financial difference between the quoted costs and actual costs. This variance is often the cause of significant lost dollars. Price, if not determined correctly, could negatively impact both the company’s and the product’s profitability. Predictive analytics can be used to support quoting activities by providing a future value based on historical job performance. The purpose of the present study is to identify whether predictive analytics can be used to predict machine rate and scrap rate to give more accuracy to quoting estimation.
Keywords: predictive analytics, flexible packaging, variable costs, production efficienc
The Cutting Edge: A Virtual Reality Simulation For Surgeons to Learn and Understand Carpal Tunnel Surgery
Get PDFAs virtual reality (VR) technology becomes increasingly popular, its applications in healthcare have gained significant traction. Many current VR training tools for surgeons lack aesthetic and functional depth; they lack an understanding of the tools needed in the operating room and have limited user experience design and interactivity. In addition, there is an opportunity for specialization with immersive training models tailored to specific procedures like carpal tunnel surgery, making it difficult for surgeons to visualize complex techniques before practicing on real patients in a realistic and guided environment. This project investigates the intersection of VR experiences and their potential to shape the future of surgical training, specifically focusing on carpal tunnel surgery, by developing an immersive experience using tools such as Figma and Cinema 4D, aiming to understand the future of user experience design for surgeons within a virtual environment. With design tools like Figma and animation software, as well as 3D rendering and animation tools such as Cinema 4D and Adobe After Effects, we can create an immersive prototype with custom UI screens, intuitive UI icons, and seamless interactions. To enhance their confidence and precision, the goal is to envision a VR environment where surgeons can gain a clearer, more engaging understanding of carpal tunnel surgery techniques before entering the operating room
End-to-End Systems Limitations in Hyperspectral Target Detection using Parametric Modeling and Subpixel Lattice Targets for Validation
Get PDFHyperspectral target detection applies algorithms to high-dimensional images to identify rare targets, or objects of interest, within cluttered scenes of diverse backgrounds. Computational demands of processing hyperspectral datasets, along with their limited availability in both public and private sectors, imposes challenges in assessing system-level limitations of detection. This research presents a methodology to quantify end-to-end sensitivities and limitations through statistical modeling across thousands of target detection scenarios. The objective is to identify specific parametric thresholds where detection begins to degrade, based on “knees” in detection curves derived from model outputs of various scenarios. The model considers subpixel targets, where an object’s signal occupies an area smaller than a pixel. Since subpixel targets are spatially unresolved, validating subpixel models is challenging with conventional datasets. To address this, a lattice-based target was developed and deployed in a UAV data collection, yielding a novel dataset with approximately 300 empirical subpixel samples for each constant fill fraction (0.2, 0.4, 0.6, 0.8). To investigate limitations, eight parameters across the imaging chain (scene, atmosphere, sensor) were considered in combination with four qualitative scenes (urban, rural, forest, desert). Broad scenario sampling of target-background combinations was guided by the Mahalanobis distance, used as a spectral contrast parameter. In total, approximately 100,000 unique scenarios were generated, with scalar detection outputs stored in a multidimensional array. These detection outputs consist of a proposed metric, the log-weighted area under the ROC curve (wAUC), which evaluates overall performance while emphasizing low false alarm rates. Results include correlation coefficients and impact scores to quantify sensitivities between system parameters and detection, revealing which parameters most influence detection under various conditions. Knees in the wAUC curves were identified using a curvature-based method, and a novel knee significance score (KSS) was introduced to rank the importance of each limitation. It concludes that aerosol visibility (2-6 km) is the leading limiter in detection, followed by the subpixel target percentage (15-35%), and background clutter, modeled using a t-distribution with 3-5 degrees of freedom. This research establishes a framework to quantify end-to-end limitations in hyperspectral target detection, extendable to additional system parameters in future studies
Back to the Basics: Extracellular Protein Interaction Networks of the Human Infant Immune System
Get PDFDatamining without an age filter proves challenging, especially when searching for direct data from human infants. Although online databases provide immune interaction networks, they often lack information about the age of data sources, resulting in categorizations as age-unspecified. This limitation underrepresents the physiology of naïve immune systems in infants, particularly since full-term infant immunity transitions to an adult-like phenotype between 24 and 30 months of age. This study aims to reconstruct an age-specific immune interaction network for full-term infants by integrating literature-based evidence with existing online database content. A list of 60 extracellular protein candidates involved in immune responses was compiled and refined based on data availability across research articles, two online databases, and cross-referenced validation. From this, 20 proteins were selected based on the strength and abundance of age-specific evidence. Search parameters were refined to evaluate both source and target interactions. Infants were categorized as pre-term or full-term, with the majority of data corresponding to the latter. The interaction networks were constructed using node (cell/protein) and edge (interaction) tables. Edge tables were organized into cell–protein and protein–protein interaction types. Every possible pairwise combination among the 20 proteins was systematically searched in the literature and carefully evaluated to confirm correct age grouping, experimental methods, and direct evidence of interaction. This study identified notable inconsistencies between protein interaction pathways presented in online databases and those reported directly in the literature. Specifically, 18 cell–protein interactions and 9 protein–protein interactions listed in age-unspecified databases were not found in studies specific to full-term infants. In contrast, literature-based findings substantially enriched the reconstructed networks: 82 of 95 cell–protein interactions and 79 of 99 protein–protein interactions were newly added, enhancing the comprehensiveness of curated data on early immune development
Predicting Patient No-Show Rate in Healthcare to Improve Operational Excellence
Get PDFThe challenge of integrating no-show predictive models into healthcare environments is complex and multifaceted, extending far beyond the distinct technical challenge of building accurate models. This thesis will discuss more on the subject of incorporating machine learning into healthcare. While algorithmic accuracy and model validation are critical, they are a mere sub-system of a much larger operational, technological, ethical, legal and human ecosystem. Predictive models face mass deployment in healthcare systems which are complex networks of stakeholders across strict regulatory frameworks and deeply embedded workflows (AlMuhaideb et al., 2019). However, for these models to result in long term and real improvements in the quality of healthcare delivery. For example, in the form of better resource allocation, decreased patient wait time, as well as increased system efficiency, their integration must be holistic and strategic. Just building a technically good model is not enough. Impact in the real world depends on how well the system fits into the existing healthcare structures, and how thoughtfully it is resolving those broader system level concerns. A significant portion of the complications of this integration arise from guaranteeing privacy and protection of patient data. Of all sensitive types of personal data, healthcare data has some of the strictest legal handling requirements, including the US Health Insurance Portability and Accountability Act (HIPAA) and the European Union\u27s General Data Protection Regulation (GDPR). Because of these laws, there are strict requirements around how patient data is collected, how it is stored, how it is accessed, and how it is shared, and for any predictive system to work with patient data these requirements must be met from the beginning. In order to integrate predictive models’ healthcare providers, they require not only to comply with these regulations but also to use strong data anonymization and encryption techniques to avoid possibilities of data breaches (AlMuhaideb et al., 2019). With sensitive patient information at stake as well as the public\u27s trust in healthcare technologies, cybersecurity measures must be robust, securing vulnerabilities at both the infrastructure and application levels. Even more important is the system interoperability issue. Significant numbers of healthcare institutions employ disparate separate electronic health record (EHR) systems, appointment scheduling platforms, and communication tools that are not well integrated and do not cooperate well with one another. With this siloed infrastructure, it then becomes very difficult to integrate in a predictive model that depends on the real- time data flow and smooth flow of patient data. For no-show prediction models to be useful, they must be incorporated into existing healthcare IT systems in a manner that facilitates real-time prediction, sends actionable outputs and is easily interpreted by clinical staff
Development of a Miniaturized Extracorporeal Membrane Oxygenation (ECMO) Device on a Microfluidic Platform
Get PDFThis research investigates the design, performance, and optimization of blood-contacting microfluidic medical devices, with each aim contributing to a broader understanding of how engineering choices influence hemocompatibility and oxygen transfer. Collectively, the findings from Aims 1A–2B offer integrated strategies to develop safer, more effective microfluidic systems for ECMO and dialysis. Aim 1A assessed hemolysis across device geometries using computational and experimental methods. All designs showed low device-induced hemolysis (\u3c 2 ppm), validating their hemocompatibility. However, Power-Law models frequently overpredicted hemolysis, indicating a need for refined parameters tailored to microfluidic flow conditions. Aim 1B demonstrated the effectiveness of thrombosis assay techniques in evaluating clot formation in microfluidic device geometries. Results showed that geometry and local shear patterns strongly influence thrombogenesis. While hemolysis remained minimal even in high-shear devices, thrombus formation was more geometry-sensitive—underscoring the need to account for both factors in design. Modifying surface properties and accounting for blood composition could further reduce thrombotic risk. In Aim 2A, we compared PDMS, polypropylene (PP), and nanoporous silicon nitride (NPSiN) membranes for oxygen transport. Findings showed that in ECMO devices, blood-side resistance, not membrane permeability, often dominates oxygen transfer. Thus, membrane choice must be paired with optimized blood-side flow for full performance gains. Aim 2B directly addressed these limitations by integrating staggered herringbone mixers (SHBs), which introduced secondary flows and enhanced mixing. The Single Channel Herringbone design yielded the highest oxygen transfer, confirming the synergy between flow geometry and oxygen transport performance. Altogether, Aims 1A and 1B establish a strong framework for hemocompatibility evaluation, while Aims 2A and 2B provide complementary insights into optimizing oxygen transfer. Their interconnections highlight the value of a systems-level approach to designing microfluidic blood-contacting devices
Hybrid Sentiment Analysis of Police Social Media: Managing Big Data Impact on Youth
Get PDFThe pervasive influence of social media has reshaped public communication, particularly for institutions like law enforcement agencies that rely on these platforms to engage with the community. As police departments increasingly utilize social media to disseminate information, interact with the public, and manage public relations, the need for effective sentiment analysis becomes crucial. Understanding how police messages are received by the public, especially by the youth, who are among the most active and impressionable users of social media, is essential for maintaining public trust and ensuring effective communication. This thesis, seeks to develop a comprehensive solution for this challenge. The primary objective of the research is to create a robust sentiment analysis model that can accurately classify the sentiments expressed in social media posts by the police, with a particular focus on fine-grained sentiment classification. This involves not only identifying whether the sentiment is positive, negative, or neutral but also capturing the nuanced emotions and opinions that can significantly influence public perception. To achieve this, the thesis proposes a hybrid approach that combines traditional rule-based methods with advanced machine learning and deep learning techniques. Rule-based methods offer the advantage of domain-specific knowledge, allowing for precise identification of sentiment-indicative phrases and words, while machine learning models, particularly those based on deep learning architectures, provide the ability to learn complex patterns from large datasets. By integrating these approaches, the proposed model aims to achieve higher accuracy and reliability in sentiment classification, even in the context of the highly varied and context-dependent language used in social media. The research also emphasizes the importance of real-time sentiment analysis. Given the rapid pace at which information spreads on social media, especially during critical incidents, a system that can monitor and analyze sentiment in real-time is crucial for law enforcement agencies. Such a system can provide immediate feedback, enabling the police to respond promptly to emerging public concerns or misinformation, thereby mitigating potential negative impacts, particularly on youth who are highly susceptible to the influence of online content. This thesis represents a significant step forward in the application of sentiment analysis to public sector communication, offering a novel approach to managing the complexities of big data in social media and addressing the critical issue of its impact on youth