SACM - United Kingdom
Permanent URI for this collectionhttps://drepo.sdl.edu.sa/handle/20.500.14154/9667
Browse
15 results
Search Results
Item Restricted An Exploration of Word Embedding Models for Phishing Email Detection(University of Southampton, 2023-09-21) Alghamdi, Rawan; Hewitt, SarahPhishing emails are dangerous cyberattacks that attackers use to steal information. Manual solutions such as blacklists can be used to detect phishing emails. However, The emergence of machine learning solutions has made phishing email detection faster and easier. This study explored and compared the performance of three deep learning models for detecting text-based phishing emails. The models used different word embedding techniques: Word2Vec, FastText, and GloVe. All three models used a Long Short-Term Memory (LSTM) classifier. Two publicly available datasets were merged to create a balanced dataset of phishing and legitimate emails using only the body text of the emails, excluding the header. The first dataset is the Fraudulent E-mail Corpus - Nigerian Letter or ”419” Fraud, which contains phishing emails. The second dataset is the Enron Email Dataset, which contains legitimate emails. The Word2Vec- LSTM model achieved the best performance, with an F1 score of 98.62% and an accuracy of 98.62%. The FastText-LSTM also performed well, but its performance was slightly lower than the Word2Vec-LSTM model, with an F1 score of 95.73% and an accuracy of 95.73%. The GloVe-LSTM model performed poorly, with an F1 score of 55.79% and an accuracy of 60.53%. We therefore conclude that using different embedding techniques with the same classifier can result in different performances for detecting and classifying phishing and legitimate emails.82 0Item Restricted The Design and Performance of Muon Scattering Tomography Reconstruction Algorithms for Applications in Nuclear Waste Identification(Saudi Digital Library, 2023-08-03) Alrheli, Ahmad F; Thompson, LeeCharacterisation of nuclear waste in the current time is well organised and each radioac- tive waste is disposed of or stored in an interim facility depending on the type of the waste. Characterised nuclear waste is expected to be documented and kept for records to be checked regularly according to the IAEA regulations. However, uncharacteristic nuclear waste is still found, especially, old nuclear waste that was stored at a time when documentation of the materials was not required. Moreover, some historical nuclear waste might contain heterogeneous contents of different types of radioactive materials. This dictates an efficient technique to resolve these issues by characterising disposed and/or stored unrecorded nuclear waste. Such techniques to investigate these nuclear waste drums without opening them are valuable to reduce the cost and the hazard of being contacted with unidentified radioactive materials. Muon Scattering Tomogra- phy (MST) technique has significantly increased in importance as a non-destructive imaging method of nuclear waste. In the past few decades, a significant amount of research has shown the efficacy of MST as an imaging method. However, there are still several areas that require further development in MST technique to contribute to nuclear waste management. An efficient imaging algorithm is requested to image and identify nuclear waste in a few hours. This thesis shows a method of optimising imaging performances of the common algorithms. This thesis shows that dividing the volume of interest by rectangular voxels with a side length of 10 mm and height of 30 mm improves the discrimination power of the imaging method. The ASR algorithm performance increased in the ability to distinguish between a 10 cm side length of uranium cube from an equally-sized lead cube with a contrast-to-noise ratio (CNR) value of 3.2 ± 0.1, compared to the CNR value of 2.2 ± 0.07 when using cubic voxels with a side length of 10 mm. Following localising high-Z materials inside nuclear waste drums, identifying these materials in a few hours is possible. It was shown that this thesis introduces two new algorithms for material classification applications which are the Hybrid (HB) and the High Angle Statistics Reconstruction (H-ASR) algorithms. It was shown that the H-ASR algorithm is able to identify 10 cm and 5 cm cubes of uranium from lead in 3 and 4.5 hours, respectively.23 0Item Restricted Analysis of Two-Dimensional Truss Bridge Design Under Uncertainty(Saudi Digital Library, 2023-08-16) Al Jumaie, Abdulaziz; Febrianto, EkyIn civil engineering, computational analysis offers invaluable insights into understanding and predicting the responses of complex structures, with truss bridges as an example. Truss bridges, characterized by their various types and applications in construction projects, are mired in uncertainties arising from factors such as load variability and environmental conditions. While traditional deterministic approaches determine their design and performance, modern analytical techniques such as Neural Networks promise enhanced comprehension of these uncertainties. This report investigates the application of Neural Networks as a surrogate model to the FEM, with application to the analysis of truss bridge configurations, specifically five key designs: Warren, Pratt, Howe, K-Truss, and Bowstring. Preliminary findings illustrate initial agreement between FEM and Neural Network predictions, with the latter's higher computational speed during the prediction phase. However, the research acknowledges certain limitations, emphasizing the importance of comprehensive analyses incorporating real-world parameters and ground-truth validations.10 0Item Restricted Machine Learning Empowered Resource Allocation for NOMA Enabled IoT Networks(Saudi Digital Library, 2023-12-20) Alajmi, Abdullah Saad; Nallanathan, ArumugamThe Internet of things (IoT) is one of the main use cases of ultra massive machine type communications (umMTC), which aims to connect large-scale short packet sensors or devices in sixth-generation (6G) systems. This rapid increase in connected devices requires efficient utilization of limited spectrum resources. To this end, non-orthogonal multiple access (NOMA) is considered a promising solution due to its potential for massive connectivity over the same time/frequency resource block (RB). The IoT users’ have the characteristics of different features such as sporadic transmission, high battery life cycle, minimum data rate requirements, and different QoS requirements. Therefore, keeping in view these characteristics, it is necessary for IoT networks with NOMA to allocate resources more appropriately and efficiently. Moreover, due to the absence of 1) learning capabilities, 2) scalability, 3) low complexity, and 4) long-term resource optimization, conventional optimization approaches are not suitable for IoT networks with time-varying communication channels and dynamic network access. This thesis provides machine learning (ML) based resource allocation methods to optimize the long-term resources for IoT users according to their characteristics and dynamic environment. First, we design a tractable framework based on model-free reinforcement learning (RL) for downlink NOMA IoT networks to allocate resources dynamically. More specifically, we use actor critic deep reinforcement learning (ACDRL) to improve the sum rate of IoT users. This model can optimize the resource allocation for different users in a dynamic and multi-cell scenario. The state space in the proposed framework is based on the three-dimensional association among multiple IoT users, multiple base stations (BSs), and multiple sub-channels. In order to find the optimal resources solution for the maximization of sum rate problem in network and explore the dynamic environment better, this work utilizes the instantaneous data rate as a reward. The proposed ACDRL algorithm is scalable and handles different network loads. The proposed ACDRL-D and ACDRL-C algorithms outperform DRL and RL in terms of convergence speed and data rate by 23.5\% and 30.3\%, respectively. Additionally, the proposed scheme provides better sum rate as compare to orthogonal multiple access (OMA). Second, similar to sum rate maximization problem, energy efficiency (EE) is a key problem, especially for applications where battery replacement is costly or difficult to replace. For example, the sensors with different QoS requirements are deployed in radioactive areas, hidden in walls, and in pressurized pipes. Therefore, for such scenarios, energy cooperation schemes are required. To maximize the EE of different IoT users, i.e., grant-free (GF) and grant-based (GB) in the network with uplink NOMA, we propose an RL based semi-centralized optimization framework. In particular, this work applied proximal policy optimization (PPO) algorithm for GB users and to optimize the EE for GF users, a multi-agent deep Q-network where used with the aid of a relay node. Numerical results demonstrate that the suggested algorithm increases the EE of GB users compared to random and fixed power allocations methods. Moreover, results shows superiority in the EE of GF users over the benchmark scheme (convex optimization). Furthermore, we show that the increase in the number of GB users has a strong correlation with the EE of both types of users. Third, we develop an efficient model-free backscatter communication (BAC) approach with simultaneously downlink and uplink NOMA system to jointly optimize the transmit power of downlink IoT users and the reflection coefficient of uplink backscatter devices using a reinforcement learning algorithm, namely, soft actor critic (SAC). With the advantage of entropy regularization, the SAC agent learns to explore and exploit the dynamic BAC-NOMA network efficiently. Numerical results unveil the superiority of the proposed algorithm over the conventional optimization approach in terms of the average sum rate of uplink backscatter devices. We show that the network with multiple downlink users obtained a higher reward for a large number of iterations. Moreover, the proposed algorithm outperforms the benchmark scheme and BAC with OMA in terms of sum rate, self-interference coefficients, noise levels, QoS requirements, and cell radii.23 0Item Restricted Predicting the Uptake of a New Medicine in England using Classification(Saudi Digital Library, 2023-12-01) Alsoghayer, Sara; Tabassum, FaizaThe National Healthcare Service (NHS) is experiencing delays of the uptake of a new medicine within their formularies, despite the National Institute for Health and Care Excellence (NICE) recommendations. Such delays not only affect pharmaceutical companies’ during the launch stage but also contribute to potential harm in patients’ health, and low global competitiveness in the life sciences sector. This study investigates the viability of predicting the speed of uptake of a new drug on a formulary level using classification algorithms. Three types of machine learning models: XGBoost, random forest, and logistic regression were employed and evaluated. The results suggest the predictive model, XGBoost, is operating on a market entry level, showing generalized predictions across various formularies. The findings also indicate there is no correlation between the formulary medicine uptake and the number of partnered organizations of a formulary, or the size of patient population.51 0Item Restricted The Effectiveness of Fcial Cues for Automatic Detection of Cognitive Impairment Using In-the-wild Data(Saudi Digital Library, 2023-11-30) Alzahrani, Fatimah; Christensen, Heidi; Maddock, SteveThe development of automatic methods for the early detection of cognitive impairment (CI) has attracted much research interest due to its crucial role in helping people get suitable treatment or care. People with CI may experience various changes in their facial cues, such as eye blink rate and head movement. This thesis aims to investigate the use of facial cues to develop an automatic system for detecting CI using in-the-wild data. Firstly, the 'in-the-wild data' term is defined, and associated challenges are identified by analysing datasets used in previous work. In-the-wild data can affect the reliability of the performance of state-of-the-art approaches. Second, this thesis investigates the automatic detection of neurodegenerative disorder, mild cognitive impairment and functional memory disorder, showing the applicability of detecting health conditions with similar symptoms. Then, a novel multiple thresholds (MTs) approach for detecting an eye blink rate feature is introduced. This approach addresses in-the-wild data challenges by generating multiple thresholds, resulting in a vector of blink rates for each participant. Then, the feasibility of this feature in detecting CI is examined. Other features considered are head turn rate, head turn statistical features, head movement statistical features and low-level features. The results show that these facial features significantly distinguish different health conditions.15 0Item Restricted IoT: current challenges and future applications(Saudi Digital Library, 2023-08-30) Almutairi, Nader; Palego, CristianoThis master's thesis explores the dynamic landscape of Internet of Things (IoT) technology, focusing on its transformative potential and obstacles. The study explores the role of IoT devices in reshaping industries, enhancing efficiency, and enhancing resource management by navigating the complex web of IoT devices. The investigation identifies key contributors to the prevalent security vulnerabilities in IoT systems and suggests strategies to strengthen their defences. The research creates a comprehensive framework that combines encryption, authentication, and anomaly detection mechanisms to address the pressing need for secure and efficient IoT solutions. Utilising cutting-edge technologies such as machine learning and blockchain, the framework not only improves the security of IoT devices, but also guarantees data integrity and user privacy. This solution's significance rests in its capacity to pave the way for safer and more reliable IoT technology, thereby nurturing confidence among users, industries, and policymakers. The results of the study demonstrate the effectiveness of the proposed framework in protecting IoT ecosystems from cyber threats, while also addressing ethical and regulatory concerns. Beyond technology, this research has implications for societal well-being, sustainable practises, and economic growth. By mitigating the security risks associated with the Internet of Things, this study establishes the foundation for realising the maximum potential of IoT technology in various industries.51 0Item Restricted Zero-Day Malware Detection using Machine Learning(Saudi Digital Library, 2023-11-08) Alqahtani, Waad; Alateef, SaadNumerous intrusion detection and prevention systems (IDPS) have been created to recognise anomalous behaviours. However, they frequently fail to detect zero-day assaults, which take use of fresh and unpatched system flaws. Zero-day attacks are a type of malicious software that take advantage of previously unknown vulnerabilities, making them challenging to identify and prevent, and inflicting significant harm to people and organisations. Due to the lack of established attack fingerprints, detecting these attacks is difficult. The conventional strategy, which depends on firewalls and intrusion detection systems (IDS) using recognised attack patterns, is easily circumvented by attackers using cutting-edge and unproven approaches. This research suggests using machine learning techniques to recognise zero-day assaults as a solution to this problem. The study will largely concentrate on unsupervised learning algorithms like clustering as well as supervised learning techniques like decision trees and support vector machines. In order to improve the effectiveness of the machine learning models in detecting zero-day malware from executable files, feature selection techniques will also be investigated. The primary goal of this research is to create an accurate and dependable approach for recognising zero-day malware so that people and organisations can defend themselves against these advanced attacks. The effectiveness of machine learning models created expressly for identifying zero-day malware from executable files was thoroughly assessed. The study shows that the Random Forest classifier demonstrated remarkable performance metrics after training. In addition, the accuracy score of the model was 0.9967, indicating a high degree of overall correct classification. Precision was 0.9956, defined as the percentage of malware samples correctly identified among all anticipated malware samples. The recall metric, which measures how many real malware samples are correctly identified as malware instances, hit 0.9978. The F1-score was determined as the harmonic mean of precision and recall. Furthermore, machine learning techniques have emerged as powerful tools in bolstering system and network security and detecting zero-day malware. This research shows that by leveraging pattern recognition and anomaly detection capabilities, machine learning models can identify potential security threats and zero-day malware attacks in real-time, enhancing the overall security posture.31 0Item Restricted THE ROLE AND IMPACT OF INFORMATION SYSTEMS AND DIGITALIZATION IN CONTAINER TERMINAL OPERATIONS(Saudi Digital Library, 2023-09-15) Alhamed, Khalid; Feng, YuanjunThis research article aims to probe the role of information system and digitalization on container operations, Artificial Intelligence (AI), and Machine Learning (ML) on container terminal operations. Ports are crucial points in supply chain and marine transportation but still data generated in this domain is largely unutilized. Although the use of machine learning methods for data-driven decision-making is prevalent in all areas, the port industry is lacking in this adoption in comparison to other segments of transportation. The purpose of this systematic literature review is to study and analyze peer-reviewed papers published in this domain to explore the role and impact of information systems on container terminal operations. This study identified a strong positive relationship between the utilization of information systems and various AI and ML technologies in improving the efficiency of container terminal operations. Although these techniques are still not utilized up to their full potential in this field, many researchers and professionals are working in the domain to plan and establish a smart port around the globe.14 0Item Restricted Enhancing Network Intrusion Detection using Hybrid Machine Learning and Deep Learning Approaches: A Comparative Analysis with the HIKARI-2021 Dataset(Saudi Digital Library, 2023-11-09) Alkhanani, Doaa; Batten, IanThis thesis presents an in-depth analysis of machine learning (ML) and deep learning (DL) methodologies for network intrusion detection, utilizing the HIKARI-2021 dataset. By leveraging models such as Random Forest, XG Boost, LSTM, and GRU, the study aimed to identify and classify malicious activities within network traffic. The models' performance was assessed primarily based on accuracy, as well as confusion matrix evaluations. Preliminary results indicate Random Forest achieved an accuracy of 93.77%, XG Boost attained 93.02%, LSTM reached 92.48%, and GRU reported 92.50%. These results were then compared to benchmark models from the literature, which achieved accuracies ranging from 98% to 99%. Through this comparative analysis, the research emphasizes the strengths, weaknesses, and the potential of each model in real-world scenarios. Notably, while the employed models showcased commendable performance, benchmark models exhibited slightly superior results, suggesting further room for model optimization and feature engineering. This research offers valuable insights into the evolving landscape of network security and sets the stage for further exploration in enhancing intrusion detection mechanisms.103 0