This paper presents a methodology for training Tacotron 2 for text-to-speech synthesis in the Kreole language, aiming to improve access to textual information for visually impaired individuals. This paper discusses Tacotron 2’s architecture, reviews relevant literature, and presents experimental implementations. By synthesizing current research trends serves as a valuable resource for TTS advancement. We trained the Tacotron 2 model to map input text sequences to Mel-spectrograms representing speech segments using a dataset of 25 audio recordings matched to their corresponding text transcripts. Segmenting audio data, using Google Colab for model training, and using HiFi-GAN as a vocoder for audio synthesis were all parts of the training process. The completeness, repeatability, and clarity of the synthesized speech were assessed in the listening assessments. Tacotron 2 performed admirably, generating intelligible sounds for 13 of 25 words despite the small sample. This study addresses the future and focuses Tacotron 2’s promise in low-resource language environments.
The emergence of fifth-generation technology (5G) has revolutionised telecommunication networks, offering enhanced mobile broadband, ultra-reliable (eMBB), ultra-low-latency communications (uRLLC), and massive machine-type communication (mMTC) service classes. This breakthrough has garnered significant attention and investment worldwide, driving innovation and growth in the digital era. However, the adoption of cloud-based 5G core (5GC) networks, while offering scalability and deployment flexibility, has posed challenges to meeting stringent latency requirements, particularly for uRLLC services specifically for robotic telesurgery. To address this problem, mobile network operators (MNOs) have turned to edge computing (EC), using the control and user plane separation (CUPS) architecture introduced in the third-generation partnership project (3GPP) release 14 specification. This architecture enables the deployment of the user plane function (UPF) closer to users, reducing latency, and improving quality of service (QoS). However, the deployment of the UPF as a standalone node on the edge of the network exposes the packet forwarding control protocol (PFCP) to cybersecurity attacks, which pose risks to telesurgery services and could even lead to loss of life. In the existing literature, only a few techniques focus on minimising these attacks when the UPF is deployed on the edge of the network far from the 5GC. Therefore, this paper reviews PFCP attacks and explores machine learning (ML) techniques to mitigate these security threats. This paper further provides recommendations and future research directions for mitigating these attacks.
With the increasingly busy lifestyles of people, advanced maintenance systems such as automated feeding has become of utmost importance to aquarium owners. Despite existing technologies, there remains a gap in smart feeding solutions tailored to the specific dietary requirements of different fish species and population sizes. This paper investigates smart fish feeding through automated detection, counting, classification and feeding of fishes within aquariums. As part of this paper, a smart population-driven aquarium maintenance system, named AquaTrack, is proposed that integrates Internet of Things (IoT), computer vision, machine learning, and sensor technologies. The system detects, classifies, and counts fish, adjusting feeding amounts based on species and population to mitigate the risks of overfeeding and underfeeding, which can significantly affect water quality. The system was evaluated through four research questions, where a general accuracy of 76.9% in fish species identification and 77.1% in fish counting were found. While promising outcomes were observed under specific conditions, variability and limitations in performance across different scenarios were also noted. Environmental factors such as lighting, reflections, shadows, and particulate matter affected the accuracies, particularly in scenarios involving smaller fish quantities. Based on these findings, the study recommends further refinement of the system to enhance its capabilities.
With the rise of the internet usage and web applications, Reflected Cross-Site Scripting (RXSS) attacks have become increasingly prevalent, accounting for over 90% of recent XSS incidents. This paper proposes a novel defense mechanism against RXSS through a browser extension, called RXSS Protect, integrated with a machine learning (ML) algorithm. This extension, compatible with Google Chrome, Microsoft Edge, and Mozilla Firefox, employs a Support Vector Machine (SVM) model to detect and block malicious scripts in real-time. The system’s architecture includes a Flask server for running the ML model, a browser extension for client-side operations, and an SQLite database for storing URL data. This approach aims to enhance web browsing security by providing an effective tool against RXSS attacks, with potential for future extensions to other types of cyber threats. The SVM model is trained on a dataset of benign and malicious URLs and XSS payloads. Evaluation focused on answering two key research questions, related to detection accuracy and performance across browsers. Results showed that RXSS Protect achieved a high accuracy of 97.53% in identifying RXSS payloads and relatively good overall performance across browsers.
In the modern contemporary world, technological advancements have revolutionized the transportation sector. The study begins with an in-depth analysis of the current traffic situation in Mauritius, and then examines existing carpooling systems in other countries. The aim of this research is to provide a more efficient and sustainable transportation solution that can ease traffic congestion in Mauritius. By utilizing state-of-the-art technology, the proposed carpooling system has the potential to significantly reduce the number of cars on the road, leading to reduced traffic congestion, decreased carbon emissions, and improved air quality. The study includes a thorough analysis of the current traffic situation in Mauritius, as well as an examination of existing carpooling systems in other countries. Additionally, the system will incorporate features such as registration, booking, SOS, user ratings, and Google Maps to ensure a positive user experience. On the implementation level of a carpooling system Flutter, Cloud Firebase, and Google Cloud Platform will be used. The proposed carpooling system will utilize real-time data to match drivers with passengers based on their location, destination, date and time. Overall, this dissertation provides a comprehensive framework for the development and implementation of a successful carpooling system in Mauritius.
In this study we investigate different strategies for enriching tweets (now posts) to improve tweets classification accuracy. In particular, we investigate the effects of emoji-to-text translation and semantic enrichment using a locally trained word2vec model. Before doing the enrichment, we develop a baseline system where we investigate the effects of stopwords removal. In our investigation, we found out that stopwords removal degrades the classification accuracy of tweets. In the first approach, we compare the difference in the classification accuracies between tweets with stopwords and tweets without stopwords. In the second approach, we enrich tweets by translating emojis contained in tweets to their equivalent text. To test the accuracy of our approaches, we employed 3 classifiers, namely, BERT classifier, Random Forest and Multinomial Naive Bayes. The third approach builds on the results of the first approach, and it uses a locally trained word2vec model to expand tweets. The results of our study suggest that semantic enrichment and translating emojis to their equivalent meaning increases the classification accuracy of tweets.
This research paper focuses on the goals of the dashboard in 5G Smart Country project 1 , the main context of which revolves around smart farming. The project focuses mainly on the utilization of applications and robots which helps to have better results in crops, serves the environment, saves time and efforts and keeps sustainability. This paper highlights the imperative, effective agriculture applications which cover almost all necessary farming steps as they have the initiative to react according to the data which are gathered from the field. It demonstrates the advantages of employing an easy-to-use dashboard in trusting the functions of UAVs (Unmanned Aerial Vehicles), field robotics UGVs (Unmanned Ground Vehicles) and sensors which are used in the technical part of the 5G project. One of the important issues that this paper discusses is the important function of the dashboard that manages the job of the applications in an easy way for the users. One of the important issues is weed detection and management. Subsequently, the paper explains the integration of AI (artificial intelligence) in the project to detect weed, to know the fertilization necessity and production quality. The conclusion includes an assessment of the current state of research, gives insight into the 5G project and provides insight into potential avenues for future research.
With advancing technologies, Extensible Markup Language (XML) has become a popular document format that is used by a wide range of applications. This document format is used for storing, exchanging, and representing data across diverse systems and platforms thereby ensuring interoperability in various applications. However, XML parsers are susceptible to a specific vulnerability known as XML External Entity (XXE) attacks. These attacks exploit weaknesses in XML processing, potentially leading to serious security breaches in systems that lack adequate security mechanisms. Whilst limited published literature has investigated XXE attacks, this study comprehensively reviews six different strategies for preventing such attacks, to eventually analyze and evaluate their effectiveness in mitigating XXE attacks. Techniques reviewed include use of XML Parser Configuration, whitelists, user input validation, static code analysis, honeypot and web application firewalls. The findings indicate that completely disabling external entities within XML parsers is the most effective approach for preventing XXE attacks. In addition, the research highlights the critical role of proper configuration and adherence to stringent security practices in XML parsing.
Being able to vote is seen as a key democratic right in many countries. Traditional means and ways of voting include the paper-based means, followed by counting. With this approach, registration procedures are sometimes difficult and the voters are expected to be physically present in the voting centers. This at times can be factors that discourage voters to vote. Electronic voting is also common but the latter can be subject many security threats, including tampering with the voting data and tampering with electronic voting systems. To overcome the issues described above, this research proposes MauVote, a novel blockchain-based voting system. The decentralized nature of blockchain eventually ensures that all transactions are immutable and recorded transparently. This greatly helps to reduce the threats of unauthorized access and alteration of data. The proposed system also makes use of OTP-based recovery and biometric authentication through a mobile app, contributing to the security of the voting process
In many post-Colonial African countries, English is the medium of instruction at schools and universities. Many, if not most of the students, however, are not home language English speakers. In the specific situation of Zimbabwe, approximately eighty percent of the population is Shona speaking and approximately ten percent of the population is Ndebele speaking. Existing research shows that when learners or students are forced to learn technical subjects (such as mathematics and computer programming) in their non-home language, they are doing double work. Not only do they need to learn the new subject domain, they also need to translate these new terms to and from their home language. This paper investigates the use of an GPT-4 based artificially intelligent tutoring bot configured to tutor the subject of Python Programming in both Shona and Ndebele. This paper is the first step of a multi-step research project which has the final goal of helping Zimbabwean learners and students learn Python Programming in their home language. This first step, however, reports on language evaluations of the GPT-4 based tutoring bot operating in Shona and Ndebele.
Voluntary engagement is an indispensable cornerstone of Critical Infrastructures (CI) such as civil protection, disaster, crisis and rescue management, health and social services, or harvests and food supply. Sustainability of the voluntary sector, however, is massively endangered by profound changes in demography, social structure, and volunteer motives in the sense of individualization and pluralization of society. This article tackles these key challenges by proposing a digital platform for goal-oriented volunteering across and independently of different non-profit organizations. This platform is intended to be a first step in synergistically aligning the goals and competencies of volunteers with the activity requirements to strengthen CI.
The advent of unmanned vehicles integrated with Vehicle-to-Everything (V2X) communication technology has revolutionized various sectors by enabling enhanced operational coordination and real-time data exchange. However, this integration has also introduced a plethora of cyber threats that pose significant risks to the safety, functionality, and security of these systems. This paper presents a comprehensive taxonomy of cyber threats targeting V2X-enabled unmanned vehicles and the corresponding countermeasures necessary for their mitigation. Through a detailed review, this study explores the technological foundations of V2X communication, its integration with different classes of unmanned vehicles, and the multifaceted cyber threats that arise from such integration. The paper further discusses the defense strategies across various domains—ground, aerial, and underwater—and emphasizes the need for robust encryption, secure authentication, and privacy-preserving techniques.
The dynamic terrain of malware attacks presents noteworthy obstacles to cyber security, necessitating for proactive and resilient detection techniques. Conventional signature-based methods struggle to keep pace with the new malware strains and obfuscation strategies. A systematic literature review was conducted to investigate existing methods for malware threat prediction and detection using machine learning and deep learning techniques. The review identified several promising approaches: convolutional neural networks, graph neural networks, and visual malware characterization achieving 95-99% accuracy on malware classification and detection tasks. However, major gaps were identified in the models’ generalizability across diverse malware types, robustness against evasion attempts, lack of interpretability due to the black-box nature of deep learning models, and limited evaluation on real-world emerging threats as opposed to controlled datasets. This project aimed to develop an AI-based threat predictive algorithm that leverages the power of deep learning and machine learning for effective malware detection and prediction. The suggested method utilises an ensemble approach that combines a convolutional neural network (CNN) for pattern recognition in malware code structures with a support vector machine (SVM) for robust decision boundaries in the feature space, thereby enhancing generalization, interpretability and adversarial resilience. By evaluating the model on the MalImg dataset, the system achieved 92.37% accuracy. Although the developed system exhibits optimal outcomes, several areas could use more improvement. This project contributes to the ongoing efforts in combating malware threats and highlights the potential of combining deep learning and traditional machine learning techniques for effective threat prediction and detection.
With the increasing sophistication of cyber threats, fileless malware has emerged as a particularly insidious form of cyberattack. Unlike traditional malware, fileless malware operates using unconventional approaches such as running entirely in memory, thus evading detection by traditional antivirus software. Whilst innovative approaches are essential to effectively detect fileless malware and mitigate potential consequences, limited studies have been conducted to examine and compare between detection approaches. To address this gap in published literature, this paper critically reviews and compares different fileless malware detection approaches. By evaluating six selected approaches, including behavioural analysis and machine learning, the study aims to provide valuable insights for enhancing cybersecurity defences against this growing threat, offering guidance for practitioners, researchers, and cybersecurity professionals.
Harmonised System (HS) code fraud, also known as Customs tariff classification fraud, is common in developed and developing countries. It occurs when traders intentionally mis-classify goods during the time of clearance. Classification fraud results in tax revenue loss, restricted and prohibited goods entry, and unfair market competition. Current manual verification techniques to detect classification fraud are time-consuming and error-prone; thus, an effective and precise strategy is required. In this study, we developed and evaluated the performance of Logistic Regression and LightGBM machine learning models in detecting HS Code fraud using a dataset obtained from a government Revenue Authority in a developing country. Results show that the LightGBM model outperformed Logistic Regression. The LightGBM model demonstrated superior accuracy (0.961), precision (0.890), recall (0.901), and F1 score (0.896) as compared to the Logistic Regression model, which had accuracy (0.932), precision (0.772), recall (0.900), and F1 score (0.831). Our contribution is threefold: Firstly, the dataset used is from a developing country, unlike other studies that use datasets from developed countries. Secondly, we worked with Customs import data collected at an 8-digit HS code level, whereas most similar studies typically used HS codes at 2, 4, or 6-digit levels. Lastly, unlike comparable studies, we have compared the performance of Logistic Regression and LightGBM machine learning algorithms.
The expression Society 5.0 was coined by the Japanese Prime Minister Shinzo Abe to describe a Society which is not driven by capital but is driven by data. Artificial Intelligence (AI) will be widely used in Society 5.0. This paper describes a project where AI was used to create a tutoring system to assist students in computer programming. The student perceptions of the tutoring bot has been previously reported. In contrast, this paper describes certain characteristics which an AI tutoring bot should have in order to help students in technical subjects in Society 5.0. Detailed examples are provided verbatim from the conversation log files recorded during the project to show how these suggested characteristics should work.
This study presents a novel formulation for the Liner Shipping Network Design Problem and proposes an Ant Colony Optimisation (ACO) based approach to solve this Np-hard optimisation challenge. The problem is formulated as a graph problem and includes key considerations such as fuel costs, idle fuel costs, operational costs, port call costs and canal costs. The proposed approach designs cost effective liner shipping networks by first segmenting the network using the Louvain algorithm and identifying key hub ports based on several factors including transhipment costs. An ACO algorithm is then applied on each subgraph to minimise the overall network costs. The performance of the proposed approach is evaluated using several scenarios based on the LINER-LIB dataset according to execution time, scalability, solution quality and robustness. The results show that the proposed approach achieves up to 66.1% cost savings on large-scale networks and reduces computational time by 65.8%, hence demonstrating its ability to optimise complex liner shipping networks effectively and efficiently.
Enhancing public understanding of flowers is essential in the endeavour to promote environmental awareness, stimulate scientific curiosity, and encourage appreciation for biodiversity. However, the lack of flower identification skills and knowledge remains a significant issue. One technology that can enable transformative visualization of flowers is Augmented Reality (AR) and this technology can potentially improve engagement and enhance botanical awareness. Nevertheless, its application in botanical education, particularly to critically explore accurate visualisation of flower species has not been well addressed in literature. This paper addresses this gap and presents an AR prototype, named Floral AR designed to enable transformative visualisation of flowers in real-time. Floral AR consists of interactive features such as 3D overlays, detailed textual content, and animations thereby depicting biological processes. The AR tool aims enhance user engagement and educational depth about flowers. To critically explore accurate visualisation, five research questions were formulated and investigated through practical application of Floral AR. The results validated the application’s robustness in accurately recognizing and tracking targets with a mean score of 82.5% in a controlled environment, confirming its viability as an educational tool.
Prioritizing competing goals among different stakeholders in requirements engineering is usually a daunting task for system architects as each of the stakeholders have their own interests. This study modeled this conflicting environment among stakeholders as a non-cooperative game in game theory with goals modeled as strategies and utility value of each goal as pay-offs in a pay-off matrix. In a user study that employed the NashPy algorithm to deduce the Nash equilibrium of the strategies in a two player game, the tool was able to automatically deduce goals which needed to be prioritized to achieve a win-win situation for two stakeholders. This was also confirmed by stakeholders. The paper has three contributions: modelling goals for stakeholders in requirements engineering as a non-cooperative game in game theory; automatically deducing goals that need prioritization among stakeholders in requirements engineering based on the Nash equilibrium; and performing a user study to validate the deduced prioritized goals.
Crown-of-thorns Starfish (COTS) outbreaks are the major direct contributors to decline in coral cover. This research investigates the application of transfer learning and YOLO variants (YOLOv5 and YOLOR) for real-time detection of Crown-of-Thorns starfish (COTS) in underwater images. Due to the scarcity of COTS-specific datasets, the potential of leveraging pre-trained models and data augmentation techniques were explored. Our findings show that models trained on close-up COTS images achieved higher accuracy compared to those trained on wider reef scenes. While YOLOv5 performed better in terms of speed and accuracy, both models struggled to detect small objects and distinguish COTS from similar marine life. To address these limitations, future research should focus on data augmentation, curriculum learning, and multi-stage detection pipelines to enhance model generalization and robustness for real-world deployment.
