22505 research outputs found
Sort by
Modeling the Effects of Shed Target Receptors on the Efficacy of Cancer Immunotherapy Agents
No full textCancer cells often shed receptors targeted by immunotherapies. Shed receptors can reduce drug efficacy by binding to free drug, preventing its binding to membrane-bound receptors. The goal of this dissertation is to investigate the effects of shed targets on the efficacy of cancer immunotherapies. First, we study liquid tumors by extending a PK/PD model to include receptor shedding, drug-induced enhancement of shedding, drug binding to shed receptors, and drug-induced tumor lysis. We use our model to elucidate the effect of shed target receptors on the efficacy of immunotherapies through uncertainty and sensitivity analyses. Our findings support the claim that the presence of shed target receptors significantly affects drug dynamics at high shed-target-to-dose ratios. Our PK/PD model is constructed as a dynamical system, and we show that its fixed point is a global attractor. Next, we consider the effect of shed receptors on the efficacy of T-cell engagers (a subtype of immunotherapies) in solid tumors by constructing a system of partial differential equations that describe the change in tumor size over time. A T-cell engager is an antibody construct that attaches to a tumor cell on one end and a T cell on the other, allowing the T cell to kill the tumor cell. We included T cells and their receptor states in our model to capture the T-cell engager dynamics. Our model provides a novel mathematical framework for T-cell engagers with and without shedding. We solved our PDE system for different shedding rates and found that increasing the shedding rate decreases the drug penetration in the tumor. The reduced drug penetration that we saw with high shedding resulted in a drastic decrease in drug efficacy. At the end of this dissertation, we discuss ways to enhance our solid tumor model and additional studies that can be conducted
Predictive Modeling of Tokenized Real Estate Prices Using Machine Learning
No full textIn this paper, we propose a machine learning-based model to predict the prices of tokenized real-estate assets, combining blockchain data, real-estate data, and sentiment data. Tokenization refers to partitioning a physical asset into fractional tokens on a blockchain, where each token represents a fraction of that asset. These include token liquidity, trading volume, platform activity, and the on-chain activity of investors. These new variables require data-based analytics that are capable of accounting for more complex relationships than customary real estate valuation approaches. To address this, we propose a multimodal dataset that incorporates on-chain data (token supply, number of wallet-holders, transaction frequency), property fundamentals (property value, property value-yield, property rent, property size), and sentiment indicators including social media analytics. We trained and evaluated four supervised machine learning models including a Neural Network, a Linear Support Vector Machine (LSVM), an XGBoost and a Linear Regression model to determine the best-performing model for predicting short-term token price movements. The Neural Network model led to the best result. The R2 equals 0.9859; the correlation coefficients for train and test datasets equal 0.982/0.982; the relative minimum error equals 0.035. The LSVM and the XGBoost based models result in similar values in terms of RMSE (0.9719 and 0.9727, consecutively). The Regression model returns the lowest value, with R2 equal to 0.628 and an error of 0.930. The findings depict the importance of using non-linear learning architectures to capture the high dimensional interdependencies between blockchain activity, property fundamentals, and investor sentiment. The study provides an understanding into the drivers of value for tokenized real estate. Feature analysis revealed that real estate value, annual yield of the asset, and wallet holders were the key predictors, with sentiment and trading volume being a secondary determinant of behavior in the tokenized real estate market. The main contribution of the research is the strong, repeatable and interpretable model that informs decision makers from the perspective of the investor, developer and regulator. This work will ultimately contribute to the transparency and efficiency of this emerging ecosystem of tokenized real estate with a thorough and extensible foundation of heterogeneous data modalities and ML-based algorithms that support scalable hybrid financial and blockchain analysis and processing of this novel real estate market segment
The Role Of Technology in Modernizing Prison Management Systems
No full textThis thesis examines the role of technology in modernizing prison management systems and improving operational efficiency, security, and decision-making. It explores how digital technologies and data analytics can address challenges in traditional prison management, such as inefficient processes and limited data utilization. The study highlights the potential of technology-driven solutions to enhance system performance and support more effective management practices within correctional facilities
Exploring Engineering Students\u27 Utilization of Resources in Calculus using Self-Regulated Learning
No full textIn response to high failure rates of engineering students in introductory math courses such as calculus, a wide variety of interventions have been implemented. A common intervention is targeted at modifying the curricula. Additionally, a major initiative to improve pass rates is to provide resources to students to help them better learn the concepts and get continual support as they complete assignments and other coursework. Despite these interventions, pass rates continue to remain low. I posit that merely the availability of resources is not enough for student success in mathematics courses. Students who lack knowledge of how to use these resources effectively may not necessarily see significant improvement in their math skills. To address this issue, in this study, I explore how engineering and engineering technology students use resources in their calculus courses. My research aims to address the gap of how students decide to use optional resources in introductory calculus courses to assist their learning and how they incorporate these resources into their learning strategies. To address this issue, I collected and analyzed interview data from 14 engineering and engineering technology students enrolled in an introductory calculus course during the Fall semester of 2024 in an R2 institution in the mid-Atlantic region of the USA. I investigated their use of resources for an introductory calculus course throughout the semester (3 interviews per student) through thematic analysis of the interviews using an adapted Self-Regulated Learning framework and Actual Student Study Paths. I found that students may choose to start, continue, drop, or refuse to use resources for any number of reasons broadly categorized as reasons related to the self (such as satisfaction with their ability), reasons related to the resource itself (such as dissatisfaction with the resource), and reasons related to external factors (such as assessment grades). The research primarily contributes to literature in two aspects. First, it extends the self-regulated learning framework to students’ use of resources to explore why students modify their use of resources. Secondly, it contributes toward understanding how students incorporate and use the resources available to them in their study habits
AI-Powered Multimodal Tour Guide: Enhancing Cultural Tourism with Image Recognition, and Personalized Storytelling
No full textThis thesis presents the design and evaluation of an AI-powered multimodal tour guide that uses image recognition and personalised storytelling to enhance cultural heritage experiences. Traditional approaches to learning about monuments rely on static plaques, generic tour content, or manual web searches, which limit personalisation, interactivity, and accessibility. To address these limitations, the study implemented a “Multimodal Monument Explorer” that allows users to upload a photo of a landmark or describe it in natural language and then receive rich, context-aware explanations in both text and audio form. The system integrates a persistent vector database of monument images, OpenCLIP-based visual embeddings, and a large multimodal language model to identify visually similar landmarks and generate tailored narratives. Multiple guide personas (e.g., historian, epic storyteller, comedian) adapt the narrative style to user preferences while preserving factual accuracy, and a Streamlit interface orchestrates retrieval, multimodal reasoning, and real-time text-to-speech synthesis. The research followed a design science and mixed- methods evaluation strategy. A quantitative study on 100 image–query pairs showed high retrieval performance (P@1 = 0.87, P@3=0.94, P@5=0.98, MRR=0.91) with end-to-end response times under eight seconds for both text and image interactions. A user study with 20 participants yielded an average System Usability Scale score of 88.5 (“Excellent”) and very positive ratings for narrative quality, engagement, and persona enhancement. The findings demonstrate the technical feasibility and user value of multimodal, persona-driven AI tour guides and offer practical design insights for future AI-driven cultural heritage applications at the intersection of data analytics, human–AI interaction, and digital tourism
Early Electrical Fault Detection in Power Systems Using Data Analytics
No full textThe increasing penetration of renewable energy, especially solar generation, has introduced higher variability into power system loading, making early detection of electrical faults more challenging yet more essential for maintaining network reliability. This research presents a machine-learning–based framework for early electrical fault detection using a Random Forest model developed in DataRobot. Due to confidentiality constraints on DEWA operational data, a Kaggle dataset was adopted and enriched with simulated solar irradiance variability to mirror real network conditions in the UAE. After comprehensive data preprocessing and feature engineering, the Random Forest model demonstrated strong generalization performance, accurately distinguishing early fault signatures from normal load fluctuations. Key findings show that the model can identify subtle precursors of faults before escalation, with performance significantly exceeding baseline models. Interpretability results from SHAP and RuleFit analyses revealed that features such as current imbalance, voltage deviations, phase-wise loading differences, and irradiance volatility were the strongest contributors to early fault prediction. Rule-based explanations indicated that specific combinations of rising current spikes and falling voltage levels consistently preceded fault events, offering actionable insights for grid operators. Overall, the study confirms that integrating machine learning into power system monitoring can meaningfully reduce missed detections, support proactive maintenance, and improve the operational reliability of distribution networks, particularly under renewable-induced variability
A Comparative Analysis of Machine Learning and Deep Learning Models for Human Activity Recognition Using Wearable Sensor Data
No full textWearable sensors have become vital for health monitoring, evaluation of sport performance, and recognition of activities of daily life in Human Activity Recognition (HAR). With the combination of machine learning and inertial sensing, new possibilities for the collection of data pertaining to human motion in daily life have emerged. Nevertheless, problems remain to be resolved in the field of modeling in relation to tradeoffs in classification. The focus remains on imaging, signal variability, and sensor noise. This study aims to determine the usefulness of Inertial Measurement Unit (IMU) data for activity recognition and compares classical machine learning methods with a temporal deep learning method. IMU and Electromyography (EMG) signals in the HuGaDB dataset, a collection of multi-sensor recordings representing a set of common daily activities, were filtered, normalized, and segmented using a sliding window, and time and frequency domain features were extracted to train Logistic Regression, kNN, Random Forest, Support Vector Machine, and MLP models. An LSTM network was trained on sequential, windowed data in tandem to capture temporal dependencies of the IMU data. The findings show that classical machine learning approaches create strong baseline performance, with Random Forest achieving the highest accuracy across models that do not employ temporal methods. The outcomes reinforce the benefits of temporal modeling for HAR from wearable sensors, while also demonstrating the benefits of simpler classical modeling in terms of interpretability and lower computational requirements. The insights from this research are useful for balanced HAR design using IMU sensors and explain the trade-offs between feature-engineered ML and deep learning. The results promote HAR systems with real-world applicability by advancing multidimensional, precise, and flexible recognition systems
ANALYZING AIRLINE CUSTOMER EXPERIENCE CHALLENGES AND THEIR IMPACT ON DUBAI\u27S TOURISM SECTOR
No full textThe paper explores whether there is a connection between operational performance, customer sentiment, and digital administrative complexity among major regional Middle Eastern airlines, such as Flag Carriers (e.g., Emirates) and Low-Cost carriers (LCCs) (e.g., Air Arabia). With the use of a highly detailed dataset of customer review and operations data, the study proves that a large service paradox is present in which, despite the high Net Promoter Scores (NPS: 48.004) of the carriers, which are positively reinforced by delivering world-class soft products, the loyalty is constantly disrupted by low-frequency but high-severe operational delays (Delay_Minutes). Since it has been analyzed that the customer friction has moved beyond the core of the flight experience to the administrative and logistical ecosystem. In the case of LCCs, this friction is concentrated in the expressions of ground logistics and visa processing, whereas in the case of Flag Carriers, this friction is fixed in the sophisticated, overarching digital infrastructure (e.g., loyalty program banking links). Moreover, sentiment models proved to be weak predictors of Flag Carriers (Emirates: 55%), which supports the notion that customers provide mixed-signal feedback-praising quality in service provision and denounce service failures at the administrative level at any given time. The paper concludes that the strategic priorities that are critical include operational reliability and simplification of digital and ground touch-points. Among the recommendations, one should single out the production of less tail-end delays and the implementation of the Aspect-Based Sentiment Analysis (ABSA) to identify the root causes of the friction and recover the services accordingly
Assessing the Impact of Codec-Induced Audio Degradation on Voice Biometric Systems
No full textThis study examines the robustness of voice biometrics when speech signals undergo audio codec transformations and sampling rate variations, conditions common in telecommunication networks. Speaker verification systems such as ECAPA-TDNN perform well on clean datasets, but their accuracy declines when low-bitrate codecs compress speech or when signals are resampled at reduced frequencies. In real-world deployments, systems adapt audio to bandwidth and storage limitations, often removing subtle acoustic details that support consistent speaker recognition. The research will analyse how codec settings and sampling rates, particularly those optimized for efficiency in bandwidth-limited systems, influence the stability of speaker embeddings. Instead of ranking codecs, the study will investigate how compression and resampling shape embedding quality and verification accuracy. To accomplish this, controlled experiments will be designed using a standardized speech dataset. Audio is systematically encoded, resampled, and decoded under multiple conditions, and the resulting signals are evaluated using cosine similarity based speaker embedding metrics, including Hit@k, mean comparisons-to-accept, and similarity regret. This study will contribute at both theoretical and practical levels. Theoretically, it will expand understanding of how bitrate and sampling distortions affect embedding behavior in modern verification systems. Practically, it will deliver recommendations for deploying voice biometrics in environments constrained by bandwidth or sampling rates. These recommendations will identify operating ranges where verification remains reliable and highlight caseswhere performance is likely to degrade. The findings will guide the development of more resilient pipelines for voice authentication in mobile and telephony applications
Harnessing AI For Assessing Government Digital Maturity
No full textThis paper provides a comprehensive analytical examination of the Global Technological Maturity Index (GTMI), with a focus on assessing the determinants of national digital maturity and structurally modeling GTMI performance. Using rigorous data preparation methods including imputation, feature transformation, and exploratory statistical analysis the study applies multiple regression models (Decision Tree, Random Forest, Gradient Boosting, and MLP) to replicate the GTMI structure and identify its primary drivers. The UAE demonstrates a strong maturity profile, ranking in the Very High category globally. Deviation analysis shows that the UAE aligns closely with global leaders across the most structurally divergent indicators. Among the models tested, the Gradient Boosting Regressor achieved the highest structural replication accuracy, confirming its suitability for modeling the GTMI architecture. Crucially, the analysis identifies CGSI as the most influential structural driver of GTMI outcomes globally, followed by PSDI and DCEI, offering evidence-based insights to support policy and strategic interventions aimed at sustaining the UAE s digital leadership