Masters Theses

Permanent URI for this collection

https://hdl.handle.net/11073/8815

Masters theses submitted by AUS graduate students.

Browse

Modeling and Control of a Robot Based Rehabilitation System for the Head-Neck Joint
(2024-06) , Ismail Tareq Raslan; Romdhane, Lotfi; Jaradat, Mohammad
A Bayesian Approach to Feature Selection in Classification Problems
(2024-07) Emarly, Maher; Alzaatreh, Ayman
The exponential growth of data, as well as the widespread use of machine learning in daily life, demonstrate the importance of feature selection. Feature selection, defined as the process of identifying and selecting a subset of relevant features from a larger set of available features, is a crucial step in machine learning. The performance and efficiency of machine learning models are improved by focusing on the most informative features and eliminating unnecessary or redundant ones. Furthermore, model interpretability is enhanced, resulting in clearer insights and an actionable understanding of the results. The resulting models are more robust, less prone to noise, and can be efficiently trained and deployed, ultimately contributing to more effective and efficient data-driven decision-making processes. We propose a Bayesian approach using the relative belief ratio (RBR) as a filter method in this paper. The proposed method showed an excellent performance in binary and multiclass classification problems. In addition, the proposed method generates a strength value that can be used as an importance score for each feature. The numerical value of the strength of the RBR is used to rank the features. This method aims to discern the relative importance of features concerning a target variable and test for their significance. The proposed method’s performance is evaluated using both synthetic and real-world datasets, and it is compared to various popular filter methods.
Outlier Detection Using the Relative Range Distribution
(2024-07) Dallah, Dania; Sulieman, Hana; Alzaatreh, Ayman
Outlier detection plays a crucial role in data analysis. Outlier detection is a challenging task due to the subjective nature of defining what constitutes an outlier. By identifying and appropriately handling outliers, analysts can gain a deeper understanding of the data, improve the quality of analyses, and make more informed decisions. In this thesis, we propose a new measure for detecting outliers in univariate data. The new measure, called relative range, is defined as the range statistic divided by the interquartile range (IQR). Since the range provides a simple yet effective measure of data dispersion, analyzing the range distribution will help identify potential outliers that fall outside the expected range of values. The probability distribution of the relative range is estimated for both symmetrical and skewed data distributions using Monte Carlo simulations. Based on the estimated empirical distribution of the relative range, a threshold is determined and used to detect potential outliers. The thesis also proposes a sequential approach for outlier detection based on the relative range. In general, the relative range has shown to be a more robust statistic at detecting outliers in both sequential and non-sequential outlier detection.
Machine Learning Approach for Predicting Appointment No-Show in Healthcare
(2024-07) Khalouf, Dana; Shamayleh, Abdulrahim; Alshariedeh, Hussam; Awad, Mahmoud
The efficiency of public healthcare delivery is essential to achieving optimal health outcomes for patients. One factor that hinders efficiency is patient no-shows, which should be managed effectively to reduce its adverse impacts on patients and healthcare providers socially and economically. In particular, it reduces patient care access, underutilizes resources, and leads to lost revenue. A no-show, or missed appointment, happens when the outpatient does not attend the scheduled appointment or cancels it at short notice. It is a common challenge faced by several healthcare systems. Previous studies have considered different models to identify patients more likely to miss their appointments; however, no study analyzed appointment no-shows in the United Arab Emirates. Therefore, this study used a data analytics and machine learning approach to develop a classification model to predict whether an outpatient will miss their appointment in Dubai's primary healthcare clinics. While data analysis is applied to extract insights from historical data and identify the most useful features, machine learning tools extrapolate on historical data to generate future predictions. A historcial dataset of appointments for the period 2021-2022 is utilized in this study. A prediction accuracy of 78% and an AUC of 0.859 were achieved using Gradient Boosted Trees while optimizing on Youden’s Index. In addition, the most influential drivers of patient no-shows were identified from the feature importances produced from the tuned model and an extensive exploratory data analysis, which included the patient’s health plan, the clinic, and the patient’s weight. As a result, recommendations of startgeies were proposed to DHA clinics to reduce no-shows, which will improve efficiency and enhance patient access to care.
Structural Performance Of 3D Printed Concrete Load Bearing Walls
(2024-06) Mohammed, Arafat Abdulrahman; Al-Tamimi, Adil
Integration of Generative Artificial Intelligence (GAI) in Academic and Engineering Sectors to Enhance Employee Productivity
(2024-08) AlNaqbi, Humaid Abdalla; Bahroun, Zied; Ahmed, Vian
Over the last several decades, the globe has seen remarkable growth in science and technology, which has resulted in fundamental advancements in a variety of areas and disciplines. This growth highlights the importance of artificial intelligence in human history, as it opens new horizons for leveraging advanced programs and technologies to enhance and increase organizational performance in a sustainable manner. However, despite this significant progress, there is still a research gap in the applications of Generative AI (GAI) in engineering and academic disciplines, as the challenges and opportunities associated with these fields have not been adequately studied. This study aims to fill this gap by investigating how GAI applications can be integrated to enhance productivity among students and faculty in the academic and engineering disciplines, which are vital sectors for the development of technological innovations. The research also addresses how to adopt this technology in a responsible and ethical manner, especially in these two important sectors. The study also included interviews and semi-structured surveys with faculty and students at a prestigious institution to explore their experiences, attitudes, and expectations regarding the use of Generative AI. Analyzing the data using the Relative Importance Index (RII) method, the results showed that compliance standards to mitigate bias were a top concern among faculty members, a point that was also confirmed by students. This study provides an important basis for future research aimed at guiding educational institutions towards effective and sustainable implementation of this technology.
Enhanced Deep Fusion Filter for Low-Cost INS/GPS Integration
(2024-04) Abdelghani, Mohamed Ismail; Jaradat, Mohammad; Abdel-Hafez, Mamoun
A Robust Approach for Enhanced Autonomous Robot Navigation
(2024-03) Ismail, Sherif Khaled Mohamed Issa; Abdel-Hafez, Mamoun
The demand for robust and precise autonomous navigation systems for mobile robots has become increasingly imperative. These systems find applications across a spectrum of industries, from logistics and transportation to agriculture and defense. Autonomous navigation offers the promise of enhanced efficiency, reduced operational costs, and a safer work environment, making it a vital component in the ever-evolving landscape of modern automation. This research aims to enhance the localization accuracy of Unmanned Ground Vehicles (UGVs) with cost-effective Global Positioning System (GPS) and Inertial Navigation System (INS) sensors by proposing a variation of the Kalman Filter (KF) for sensor fusion. This algorithm is used in this work for a low-cost loosely coupled GPS/INS integration. To achieve this objective, we propose the utilization of the Variational Bayesian-based Maximum Correntropy Cubature Kalman filter (VBMCCKF). This filtering technique demonstrates exceptional adaptability and resilience, making it particularly well-suited for the dynamic and often unpredictable environments that autonomous vehicles must navigate. The research strategy encompasses a dual-phase approach. Initially, a comprehensive evaluation of the performance of EKF, CKF, VBCKF, and VBMCCKF algorithms is conducted in a distinct application, namely Li-ion battery estimation. This preliminary phase is aimed at establishing the efficacy of these algorithms. Subsequently, in the core application, an offline estimation will be carried out using experimentally derived vehicle trajectories to ensure meticulous accuracy verification. It is noteworthy that the VBCKF as well as the VBMCCKF are being applied in this particular application for the first time. Following this, a transition was made towards real-time deployment for the localization of the autonomous navigation of a UGV. Throughout this phase, a thorough assessment of practicality and resilience are undertaken in dependable operational scenarios.
Control of Unmanned Aerial Manipulator
(2024-07) Mahfouz, Saad Eddin; Jaradat, Mohammad; Mukhopadhyay, Shayok
Intelligent Rapidly-Exploring Random Tree Star Algorithm
(2024-05) Ahmed, Khidir Galal Eldin Khidir; Jaradat, Mohammad; Romdhane, Lotfi
Autonomous robots have been increasingly employed to supplant human labor across diverse fields over recent decades, serving as a foundational element in numerous industries ranging from supply chains and assembly lines to transportation. In these sectors, rapid and efficient operation is indispensable. Therefore, the development of advanced path planning techniques implies pivotal importance to mitigate human dependency. Hence, in this work, we developed an improved path planning algorithm inspired by the directional implementation in Rapidly-Exploring Random Tree Star Normal (RRT*N) and its variants, which is used to address the lack of environment adaptability and the improvement of path quality and inadequate long processing times. This new method is called Neural Adaptive Rapidly-Exploring Random Tree Star Normal (NA-RRT*N). The advanced presented method can deal with path planning problems in 2D and 3D environments. This novel method uses a Gaussian probability distribution with variable standard deviation to generate new nodes, which is controlled via Artificial Neural Network based on the environmental feedback. This feature results in a varied tree concentration in the direction of the target. It is shown that this method can be more than 68% faster in finding the initial path to the target and produces at least 5% shorter path in worst case scenario compared to three states of the art versions of RRT method. Furthermore, NA-RRT*N stood out with a perfect 100% success rate in all seven 2D scenarios tests while continually improving path smoothness. For instance, in 100 trials of the presented static scenarios, NA-RRT*N exhibited the shortest average processing time and path length across seven varied complexity maps.
Electric Scooter Battery Management and Battery Swapping using Robotic Arm
(2024-03) Daoud, Abeer Mazen; Romdhane, Lotfi; Rehman, Habib-Ur
Threshold Functions for Modeling Gene Regulatory Networks
(2024-06) Kittaneh, Hadeel Ali; Jarrah, Abdul Salam
In this thesis, we explore the properties of threshold functions with a unique updating rule and their relevance to gene regulatory models. We conduct a comparative analysis between Threshold Boolean Networks (TBNs) and Random Boolean Networks, focusing on variations in the number of inputs per gene. This analysis helps us understand how input connectivity influences network stability and phase transitions. We also investigate the dynamics and robustness of TBNs, emphasizing fixed interaction rules characteristic of genetic systems, unlike previous studies that use annealed approximations. Furthermore, we propose a new approach to assess the robustness of these networks, addressing the issue of multiple attractors in threshold Boolean networks. Our findings enhance the understanding of threshold Boolean functions and their applications in modeling gene regulatory networks.
Characterization of the dynamic flow response in microfluidic devices
(2024-05) Elgack, Mohammed Elmahdi; Abdelgawad, Mohamed
The impact of socio-economic factors on waste generation in the UAE
(2024-04) Daoud, Omar Walid Omar; Ahmed, Vian
EEG-based brain source localization of mental stress using the SAFFIRE method
(2024-04) Zahour, Nada; Al-Nashash, Hasan; Mir, Hasan
Deepfakes Recognition with Physiological Signals
(2024-04) Khan, Muhammad Riyyan; Tariq, Usman; Al-Nashash, Hasan; Dhall, Abhinav
Environmental Sustainability Assessments on Water Distribution Networks: A Life-Cycle Approach
(2024-05) Alshaikh, Abdulaziz Salah; Mortula, Maruf
Water Distribution Networks (WDNs) are considered one of the vital aspects in an infrastructure. Their significance in supplying good quality water to the end user is undeniable. Nevertheless, the processes, methods, and materials used to create WDNs are contributing towards increased environmental impacts, threatening the sustainability of the entire water supply system. Depending on its operation and maintenance throughout the life cycle, the environmental impact can be quite significant, threating its functionality, and shortening its service life. The thesis proposes a holistic approach to quantify the environmental sustainability of four different WDNs which are the Grid, Ring, Radial, and Tree WDNs using a life-cycle assessment (LCA) approach. The LCA was done using Simapro 8. The LCA is aimed at quantifying seven different environmental sustainability indicators in the form of Global Warming Potential (GWP), Ozone Layer Depletion (OLD), Cumulative Energy Demand (CED), Air Pollution and Water Pollution, Eutrophication Potential (EP), and Acidification Potential (AP) of each WDN. This will be assessed against each of the material extraction, manufacturing, transportation, and disposal life-cycle phases. The above-mentioned parameters were used to formulate the Environmental Sustainability Model and rank the four WDNs. The model was created to function beyond the reach of the study to investigate the environmental sustainability of any piping material relative to those examined. The results portrayed that the material extraction phase has the greatest influence and emissions in the LCA for each WDN. In addition, the Grid WDN is the most sustainable with an Environmental Sustainability Index (ESI) of 23.9 compared to 24.9 for Tree, 25.0 for Ring, and 25.1 for Radial.
Data-Driven Electricity Demand Modeling for Electric Vehicles Using Machine Learning
(2024-07) Far, Reza Davoodi; Hassan, Mohamed; Osman, Ahmed
Hand-Crafted Features with Simple Deep Learning Architectures for Human Activity Recognition
(2024-06) Albadawi, Yaman Sufian; Shanableh, Tamer
With the growth in the wearable device market, wearable sensor-based human activity recognition systems have been gaining increasing interest in research because of their rising demands in many areas. This research presents a novel sensor-based human activity recognition system that utilizes a hand-crafted feature extraction technique associated with a deep learning method for classification. In this work, we divide the sensor sequences time-wise into non-overlapping 2D segments. We then compute statistical features from each 2D segment using two approaches; the first approach computes features from the raw sensor readings, while the second approach applies time-series differencing to sensor readings prior to feature calculations. Applying time-series differencing to 2D segments helps identify the underlying structure and dynamics of the sensor reading across time. We also experiment with two selection methods, including stepwise regression and selecting KBest to select useful features in an attempt to create a more representative model of the extracted features. Also, we investigate the effect of adding a one-dimensional convolutional layer and an attention layer to the deep learning network on the model performance. We experiment with different numbers of 2D segments of sensor reading sequences. We also report results with and without the use of different components of the proposed system. The proposed feature extraction method is integrated with an existing transformer designed for human activity recognition. All of these arrangements are tested with different deep-learning architectures. Several experiments are performed on four benchmark datasets: mHealth, USC-HAD, UCI-HAR, and DSA. The experimental results revealed that the proposed system outperforms the human activity recognition rates and F1-scores reported in the most recent studies. Specifically, we report recognition rates of 99.17%, 81.07%, 99.44%, and 94.03% for the four datasets, respectively.
Optimizing Field Programmable Gate Array Resources in The Cloud Computing Infrastructure: Partitioning and Resource Allocation
(2024-06) Abdalgawad, Nada Mohamed; Aburukba, Raafat; Sagahyroon, Assim

Browse

Recent Submissions