Saudi Cultural Missions Theses & Dissertations
Permanent URI for this communityhttps://drepo.sdl.edu.sa/handle/20.500.14154/10
Browse
50 results
Search Results
Item Restricted Enhancing Breast Cancer Diagnosis with ResNet50 Models: A Comparative Study of Dropout Regularization and Early Stopping Techniques(University of Exeter, 2024-09-20) Basager, Raghed Tariq Ahmed; Kelson, Mark; Rowland, SarehEarly detection and treatment of breast cancer depend on accurate image analysis. Deep learning models, particularly Convolutional Neural Networks (CNNs), have proven highly effective in automating this critical diagnostic process. While prior studies have explored CNN architectures [1, 2], there is a growing need to understand the role of dropout regularization and fine-tuning strategies in optimizing these models. This research seeks to improve breast cancer diagnosis by evaluating ResNet50 models trained from scratch and fine-tuned, with and without dropout regularization, using both original and augmented datasets. Assumptions and Limitations: This research assumes that the Kaggle Histopathologic Cancer Detection dataset is representative of real-world clinical images. Limitations include dataset diversity and computational resources, which may affect generalization to broader clinical applications. ResNet50 models were trained on the Kaggle Histopathologic Cancer Detection dataset with various configurations of dropout, early stopping, and data augmentation [3–6]. Performance was assessed using accuracy, precision, recall, F1-score, and AUC-ROC metrics [7, 8]. The best-performing model was a ResNet50 trained from scratch without dropout regularization, achieving a validation accuracy of 97.19%, precision of 96.20%, recall of 96.90%, F1-score of 96.55%, and an AUC-ROC of 0.97. Grad-CAM visualizations offered insights into the model’s decision-making process, enhancing interpretability crucial for clinical use [9,10]. Misclassification analysis showed that data augmentation notably improved classification accuracy, particularly by correcting previously misclassified images [11]. These findings highlight that training ResNet50 without dropout, combined with data augmentation, significantly enhances diagnostic accuracy from histopathological images. Original Contributions: This research offers novel insights by demonstrating that a ResNet50 model without dropout regularization, trained from scratch and with advanced data augmentation techniques, can achieve high diagnostic accuracy and interpretability, paving the way for more reliable AI-powered diagnostics.6 0Item Restricted Quantifying and Profiling Echo Chambers on Social Media(Arizona State University, 2024) Alatawi, Faisal; Liu, Huan; Sen, Arunabha; Davulcu, Hasan; Shu, KaiEcho chambers on social media have become a critical focus in the study of online behavior and public discourse. These environments, characterized by the ideological homogeneity of users and limited exposure to opposing viewpoints, contribute to polarization, the spread of misinformation, and the entrenchment of biases. While significant research has been devoted to proving the existence of echo chambers, less attention has been given to understanding their internal dynamics. This dissertation addresses this gap by developing novel methodologies for quantifying and profiling echo chambers, with the goal of providing deeper insights into how these communities function and how they can be measured. The first core contribution of this work is the introduction of the Echo Chamber Score (ECS), a new metric for measuring the degree of ideological segregation in social media interaction networks. The ECS captures both the cohesion within communities and the separation between them, offering a more nuanced approach to assessing polarization. By using a self-supervised Graph Auto-Encoder (EchoGAE), the ECS bypasses the need for explicit ideological labeling, instead embedding users based on their interactions and linguistic patterns. The second contribution is a Heterogeneous Information Network (HIN)-based framework for profiling echo chambers. This framework integrates social and linguistic features, allowing for a comprehensive analysis of the relationships between users, topics, and language within echo chambers. By combining community detection, topic modeling, and language analysis, the profiling method reveals how discourse and group behavior reinforce ideological boundaries. Through the application of these methods to real-world social media datasets, this dissertation demonstrates their effectiveness in identifying polarized communities and profiling their internal discourse. The findings highlight how linguistic homophily and social identity theory shape echo chambers and contribute to polarization. Overall, this research advances the understanding of echo chambers by moving beyond detection to explore their structural and linguistic complexities, offering new tools for measuring and addressing polarization on social media platforms.7 0Item Restricted Adaptive encryption scheme for IoT sensors network(Cardiff University, 2024-09-05) Almalki, Ohud; Li, ShancangArtificial Intelligence (AI) and the Internet of Things (IoT) have revolutionised the way we live and work, bringing unpredictable levels of automation and decision-making. As a result, industries such as healthcare, finance, and smart cities have experienced significant changes. These technologies have transformed our lives to be more efficient, convenient, and connected. However, the rapid advancement of AI and IoT has also raised some concerns. Data privacy and security have become a major challenge with these systems processing massive amounts of sensitive personal and organisational information data. Highlighting the importance of implementing robust protection methods. This dissertation focuses on the different techniques used to maintain data privacy in AI and IoT ecosystems using privacy-preserving technologies (PETs), such as differential privacy (DP), federated learning (FL), and secure computation. These technologies are essential for compliance with data protection regulations such as the General Data Protection Regulation (GDPR) and the California Consumer Privacy Act (CCPA). Moreover, it is important to educate users about the associated risks of using AI and IoT and to encourage responsible behaviours. The core focus of this research is a dual-layer encryption schema that helps to protect sensitive data in IoT sensor networks by classifying the data as low and high-critical.5 0Item Restricted Pixels to Pavements(University College London, 2024) Hababi, Abdullah; Selby, EllyThe convergence of machine learning (ML) and the built environment is redefining traditional design decision-making processes. This report explores the integration of ML within architecture, urban design, and urban planning, emphasizing its transformative potential as a design decision making tool. The report delves into the historical context of digital tools in architecture and examines how ML is currently utilized in the built environment. Through a detailed methodology, the report analyzes ML’s role as a computational design aid, as a design facilitator or augmenter, and as a co-designer. This report aims to connect the idea of machine learning’s use in design decision-making processes in the built Environment to my design project. The impact of a literature review and case studies has helped extract and implement different key methods of machine learning in various stages of my design project, such as the data manipulation stage, form finding stage, design intervention placement stage, and simulation analysis of and for design decisions stage. Critical analyses focus on the role of data quality, human agency, and the limitations of ML, such as algorithmic bias and the potential erosion of human creativity. This report contends that ML can profoundly influence and effectively dictate design decision making in both an architectural and urban design context, through its aid as a computational design tool, design facilitator, and co-designer. The discussion emphasizes the necessity of human expertise in interpreting ML outputs and proposes a collaborative approach between human intuition and ML capabilities. The report concludes by advocating for a continuous dialogue between technology and human creativity to ensure ML serves as a valuable tool in shaping the built environment rather than a replacement for human ingenuity.6 0Item Restricted Supervised Machine Learning. A Strategic Approach for Financial Fraud Detection(University of Nottingham, 2024-03) Bashehab, Omar Sami; Wang, HuamaoFinancial fraud is an increasingly concerning issue in the present day. The rapidly growing rate of fraudulent activities has led to significant financial losses for many stakeholders. Card-not-present (CNP) fraud has risen with the growth of digital sales. The same benefits attracting online banking and transactions have attracted fraudsters and cybercriminals. Consequently, the incentive for fraud detection for mitigating financial risk is evident. However, traditional detectors are outdated, and rule-based systems fail to keep up with the dynamic innovative methodologies of cybercriminals. Thus, the ML-based system is needed. However, various challenges exist within ML-based detectors. Firstly, datasets are typically highly imbalanced and secondly, a lack of real-world datasets makes research extremely difficult. To tackle these problems, different resampling methods such as RUS, ROS, SMOTE and a hybrid sampling approach (ROS + RUS) were used and evaluated. Furthermore, a novel dataset was used, augmented from an original PAYsim real-world synthetic data. Furthermore, predictive models such as Decision Tree, Logistic Regression, Random Forests, Support Vector Machine and (Gaussian) Naïve Bayes were used with the different resampling methods in a comparative approach. Finally, the importance of data preprocessing and feature engineering was explored and evaluated amongst the classifiers. The experimental results illustrate the Random Forest, with Grid Search CV optimisation and RUS as well as feature engineering performed the best. The methodological approach exhibited an increase in F1 score, True Positive Rate, Recall and Accuracy for the classifier. The final model outputted an F1 score of 69%, ROC-AUC of 88% and True Positive Rate (TPR) of 93%.59 0Item Restricted LIGHTREFINENET-SFMLEARNER: SEMI-SUPERVISED VISUAL DEPTH, EGO-MOTION AND SEMANTIC MAPPING(Newcastle University, 2024) Alshadadi, Abdullah Turki; Holder, ChrisThe advancement of autonomous vehicles has garnered significant attention, particularly in the development of complex software stacks that enable navigation, decision-making, and planning. Among these, the Perception [1] component is critical, allowing vehicles to understand their surroundings and maintain localisation. Simultaneous Localisation and Mapping (SLAM) plays a key role by enabling vehicles to map unknown environments while tracking their positions. Historically, SLAM has relied on heuristic techniques, but with the advent of the "Perception Age," [2] research has shifted towards more robust, high-level environmental awareness driven by advancements in computer vision and deep learning. In this context, MLRefineNet [3] has demonstrated superior robustness and faster convergence in supervised learning tasks. However, despite its improvements, MLRefineNet struggled to fully converge within 200 epochs when integrated into SfmLearner. Nevertheless, clear improvements were observed with each epoch, indicating its potential for enhancing performance. SfmLearner [4] is a state-of-the-art deep learning model for visual odometry, known for its competitive depth and pose estimation. However, it lacks high-level understanding of the environment, which is essential for comprehensive perception in autonomous systems. This paper addresses this limitation by introducing a multi-modal shared encoder-decoder architecture that integrates both semantic segmentation and depth estimation. The inclusion of high-level environmental understanding not only enhances scene interpretation—such as identifying roads, vehicles, and pedestrians—but also improves the depth estimation of SfmLearner. This multi-task learning approach strengthens the model’s overall robustness, marking a significant step forward in the development of autonomous vehicle perception systems.33 0Item Restricted Deep Learning Approaches for Multivariate Time Series: Advances in Feature Selection, Classification, and Forecasting(New Mexico State University, 2024) Alshammari, Khaznah Raghyan; Tran, Son; Hamdi, Shah MuhammadIn this work, we present the latest developments and advancements in the machine learning-based prediction and feature selection of multivariate time series (MVTS) data. MVTS data, which involves multiple interrelated time series, presents significant challenges due to its high dimensionality, complex temporal dependencies, and inter-variable relationships. These challenges are critical in domains such as space weather prediction, environmental monitoring, healthcare, sensor networks, and finance. Our research addresses these challenges by developing and implementing advanced machine-learning algorithms specifically designed for MVTS data. We introduce innovative methodologies that focus on three key areas: feature selection, classification, and forecasting. Our contributions include the development of deep learning models, such as Long Short-Term Memory (LSTM) networks and Transformer-based architectures, which are optimized to capture and model complex temporal and inter-parameter dependencies in MVTS data. Additionally, we propose a novel feature selection framework that gradually identifies the most relevant variables, enhancing model interpretability and predictive accuracy. Through extensive experimentation and validation, we demonstrate the superior performance of our approaches compared to existing methods. The results highlight the practical applicability of our solutions, providing valuable tools and insights for researchers and practitioners working with high-dimensional time series data. This work advances the state of the art in MVTS analysis, offering robust methodologies that address both theoretical and practical challenges in this field.15 0Item Restricted Utilizing Artificial Intelligence to Develop Machine Learning Techniques for Enhancing Academic Performance and Education Delivery(University of Technology Sydney, 2024) Allotaibi, Sultan; Alnajjar, HusamArtificial Intelligence (AI) and particularly the related sub-discipline of Machine Learning (ML), have impacted many industries, and the education industry is no exception because of its high-level data handling capacities. This paper discusses the various AI technologies coupled with ML models that enhance learners' performance and the delivery of education systems. The research aims to help solve the current problems of the growing need for individualized education interventions arising from student needs, high dropout rates and fluctuating academic performance. AI and ML can then analyze large data sets to recognize students who are at risk academically, gauge course completion and learning retention rates, and suggest interventions to students who may require them. The study occurs in a growing Computer-Enhanced Learning (CED) environment characterized by elearning, blended learning, and intelligent tutelage. These technologies present innovative concepts to enhance administrative procedures, deliver individualized tutorials, and capture students' attention. Using predictive analytics and intelligent tutors, AI tools can bring real-time student data into the classroom so that educators can enhance the yields by reducing dropout rates while increasing performance. Not only does this research illustrate the current hope and promise of AI/ML in the context of education, but it also includes relevant problems that arise in data privacy and ethics, as well as technology equality. To eliminate the social imbalance in its use, the study seeks to build efficient and accountable AI models and architectures to make these available to all students as a foundation of practical education. The students’ ideas also indicate that to prepare the learning environments of schools for further changes, it is necessary to increase the use of AI/ML in learning processes15 0Item Restricted Enhancing Network Security through Machine Learning and Threat Intelligence Integration in Next-Generation Firewall IDS/IPS Systems(Northumbria University, 2024-09-05) Sufi, Mohammed; Abosata, NassrThis dissertation explores how Machine Learning (ML) and real-time Threat Intelligence feeds can improve Next-Generation Firewall (NGFW) systems especially in increasing the accuracy and efficacy of Intrusion Detection and Prevention Systems which contribute in enhancing network security. Using threat intelligence feeds including IP addresses, domains, and URLs which come with related information’s such as the Indicators of Compromise (IoC) reputation scores, and threat categories like "malware" or "phishing,”. Thus, by using this information, applying supervised learning techniques enable to easily assess and classify threats into high-risk and low risk categories in order to reduce false positives, which result in enhancing threat detection and prevention accuracy. These classified threat feeds are dynamically updated, allowing the NGFW to protect against new threats by adjusting its security rules with appropriate countermeasures. The results show that combining ML with classified threat feeds improves the NGFW's capacity to detect and prevent threats, leading to more focused and responsive threat management.22 0Item Restricted Forecasting OPEC Basket Oil Price and Its Volatilities Using LSTM(University College London, 2024-09) Almazyad, Sulaiman; Hamadeh, LamaThe global economy is greatly affected by oil prices, which have an impact on everything from consumer goods prices to transportation expenses. Forecasting these prices accurately is crucial for energy security, company strategy, and economic planning. Traditional statistical models such as ARIMA and SARIMA have been used for such forecasts, but struggle with the non-linear patterns inherent in oil price movements. This research explores the use of Long Short-Term Memory (LSTM) networks, a specialized form of Recurrent Neural Network (RNN) built to manage longterm dependencies, in predicting the OPEC reference basket oil price and its associated volatility, ultimately improving the accuracy of these forecasts. The model is built upon historical datasets of the OPEC Reference Basket (ORB), and its efficacy is assessed using a variety of performance indicators, including RMSE, MAE, and MAPE. The outcomes reveal that the LSTM model is12 0