Search

Purpose Cities have recognised information and communication technology (ICT) as a major enabler for transforming into “smarter” service providers. Mauritius is considering embarking on a smart island initiative, and as such, the existing towns in Mauritius have a major role to play in the transformation process. This study aims to focus on a systematic analysis of ICT architectures. This study proposes recommendations with regard to ICT infrastructures. This study elicits the contribution of those services towards the overall transformation of the towns, which in turn will form an integral part of the smart island initiative of the country. Design/methodology/approach A first step towards developing a smart ICT architecture is to understand what a smart ICT architecture constitutes of and the frameworks that can be used to develop these architectures. The main objectives of this systematic literature review (SLR) are as follows: understand what a smart ICT architecture consists of and derive a definition that will be used throughout the research for the development of ICT architecture for the smart local councils for Mauritius; analyse various frameworks, architectures and models that have been used in existing literature for the development of smart ICT architectures. Findings This SLR was conducted with a view to investigate ICT architectures with respect to smart governance. An in-depth review, followed by a critical analysis was performed based on the research questions set. The contribution of this study include the formulation of a definition for ICT architecture for smart local councils, which can be applied to any sphere characterised by similar aspects with reference to smart local governance. Albeit existing frameworks such as The Open Group Architecture Framework, Federal Enterprise Architecture and Zachman Framework, it was observed that in most cases, ICT architectures were built from scratch potentially because of specificities of each country/city. Originality/value This work proposes ICT architectures for smart local council.

The lack of annotated datasets makes the automatic detection of skin problems very difficult, which is also the case for most other medical applications. The outstanding results achieved by deep learning techniques in developing such applications have improved the diagnostic accuracy. Nevertheless, the performance of these models is heavily dependent on the volume of labelled data used for training, which is unfortunately not available. To address this problem, traditional data augmentation is usually adopted. Recently, the emergence of a generative adversarial network (GAN) seems a more plausible solution, where synthetic images are generated. In this work, we have developed a deep generative adversarial network (DGAN) multi-class classifier, which can generate skin problem images by learning the true data distribution from the available images. Unlike the usual two-class classifier, we have developed a multi-class solution, and to address the class-imbalanced dataset, we have taken images from different datasets available online. One main challenge faced during our development is mainly to improve the stability of the DGAN model during the training phase. To analyse the performance of GAN, we have developed two CNN models in parallel based on the architecture of ResNet50 and VGG16 by augmenting the training datasets using the traditional rotation, flipping, and scaling methods. We have used both labelled and unlabelled data for testing to test the models. DGAN has outperformed the conventional data augmentation by achieving a performance of 91.1% for the unlabelled dataset and 92.3% for the labelled dataset. On the contrary, CNN models with data augmentation have achieved a performance of up to 70.8% for the unlabelled dataset. The outcome of our DGAN confirms the ability of the model to learn from unlabelled datasets and yet produce a good diagnosis result.

In this paper, we have deployed an application for the automatic detection and classification of vehicle damages to be used by the insurance companies for processing claims or by the police department for the recording of accidents. It is often tedious to manually identify the types and severity of damages on vehicles occurred after an accident. Automated damage detection application can ease the insurance claim process. Convolution Neural Networks (CNN) have achieved great success in the classifications of objects. However, CNN has not been fully explored and applied for the multiclass classifications of vehicle damages. In this work, we have adapted the pre-trained CNN models namely the MobileNet and VGG19 and applied a transfer learning on our large constructed dataset. We have also investigated overfitting and have trained the model to learn more general features. Our application has achieved an accuracy of 70% for MobileNet and 50% for VGG19. Unlike existing models, our proposed model is able to detect and classify multiple types of damages namely dents, scratches, broken and not damage, which is an achievement in this problem domain.

Enterprise Resource Planning (ERP) systems are among the most popular IT systems and skills related to ERP implementations are key for industry. However, studies focusing on specific ERP-related skills are scarce. The main purpose of this work is to analyze skills required for ERP implementation projects. Having better insights on the types of skills required in industry can aid in identifying relevant training needs and ultimately help in capacity building for ERP related projects. A systematic review was adopted as methodology for literature analysis. A list of papers focusing on ERP skills were retrieved and were analyzed before deriving a comprehensive list of skills categorized under different themes. The findings of this paper provide a detailed list of ERP skills which can be used by industry to define training needs and by educational practitioners when designing ERP-related curriculum.

In this work, an enhanced ResNet deep learning network, depth-ResNet, has been developed to classify the five types of Tuberculosis (TB) lung CT images. Depth-ResNet takes 3D CT images as a whole and processes the volumatic blocks along depth directions. It builds on the ResNet-50 model to obtain 2D features on each frame and injects depth information at each process block. As a result, the averaged accuracy for classification is 71.60% for depth-ResNet and 68.59% for ResNet. The datasets are collected from the ImageCLEF 2018 competition with 1008 training data in total, where the top reported accuracy was 42.27%.

Mauritius stands as one of the few countries in the world to have controlled the current pandemic, the novel coronavirus 2019 (COVID-19) to a significant extent in a relatively short lapse of time. Owing to uncertainties and crisis amid the pandemic, as an emergency announcement, the World Health Organization (WHO) solicits the help of health authorities, especially, researchers to conduct in-depth research on the evolution and treatment of COVID-19. This paper proposes an integer-valued time series model to analyze the series of COVID-19 cases in Mauritius wherein the corresponding innovation term accommodates for covariate specification. In this set-up, sanitary curfew followed by sanitization and sensitization campaigns, time factor and safe shopping guidelines have been tested as the most significant variables, unlike climatic conditions. The over-dispersion estimates and the serial auto-correlation parameter are also statistically significant. This study also confirms the presence of some unobservable effects like the pathological genesis of the novel coronavirus and environmental factors which contribute to rapid propagation of the zoonotic virus in the community. Based on the proposed COM-Poisson mixture models, we could predict the number of COVID-19 cases in Mauritius. The forecasting results provide satisfactory mean squared errors. Such findings will subsequently encourage the policymakers to implement strict precautionary measures in terms of constant upgrading of the current health care and wellness system and re-enforcement of sanitary obligations.

Biometrics are being implemented to overcome the major security concerns faced by the traditional security systems such as passwords and pincode. However, over the past few years, many researchers have deployed unimodal biometrics, which makes use of only one biometric trait. Despite the fact that unimodal biometrics have solved many security issues, they are experiencing several challenges such high error rates, noisy data and intra-class variations. To address these issues, researchers are exploring multimodal biometrics, which makes use of more than one biometric trait. Despite ongoing researches, the most appropriate level at which fusion has to be done in multibiometric has still not been established. In this work, the four levels of fusions, that is, sensor level, feature level fusion, score level and decision level are being explored and experimented on various datasets using face biometrics. The performance at each level is being evaluated and the results reveal that score level fusion beats the other levels of fusion by providing a recognition rate of nearly 98.9.

In this digital era, the focus is now on the development of applications that allow human beings and machines to interact directly. Up to now, many hand gesture recognition systems have been developed for different applications such as sign language recognition and smart surveillance. In recent years, researchers have shown interest in the development of hand gesture recognition applications for dancing movements, which involve dynamic hand gestures. However, there are still various challenges such as extraction of invariant factors, automatic segmentation, the transition between gestures, mixed gestures issues, nature of dynamic hand gestures and occlusions that need to be addressed. This research work aims at developing an application to categorise and recognise the classical “Bharatanatyam” dance hand gestures. Since no online database of “Bharatanatyam” gestures is available to the public for research purposes, a customised database has been built with 900 images, consisting of 15 instances for each hand gesture. In this work, Chain Codes and Histogram of Oriented Gradients (HOG) are proposed for the feature representation of the hand gestures. For the classification of the images, Support Vector Machine (SVM) and K-Nearest Neighbor (KNN) are explored. From the experiments conducted, Chain codes with SVM provide a recognition rate of 99.9% and a false rejection rate of only 0.1%, which is a promising technique for the deployment of movement recognition applications.

Surveillance system used for monitoring existed since decades. However, these surveillance systems required human intervention for the monitoring of suspicious behaviors. With the rapid revolution in technology, automatic detection of uncommon behaviors is gaining much attention from researchers. Being rapidly accepted in most public places to ensure transparency and security, surveillance systems are contributing in many applications like live traffic monitoring, crime scenes, and old people care. The deployment of automatic detection is very complex since it requires complex algorithms that should accurately detect uncommon behaviors, which is context-sensitive. In addition, it involves a lot of incoming data from various cameras, making it more challenging. In this perspective, the factors affecting surveillance systems and the techniques devised so far to detect uncommon behavior from these systems are analyzed and discussed. Robust automatic detection applications may yield to proactive decisions to be taken to prevent any harms/damages that would be caused by any uncommon/suspicious behaviors. Thus, it is important to explore techniques that can be used to implement automatic surveillance systems.

In this paper, the machine learning algorithms have been applied on distinct features of Keystroke Dynamics. The Machine learning is important to correctly authenticate an individual. In this work, the complex models and algorithms help to determine when the person is a genuine user or an imposter through learning. The algorithms that has been studied and deployed,are the Fuzzy Expert System (FESs), NeuroEvolution of the augmenting topology (NEAT), Proposed NeuroEvolution of the augmenting topology, Support Vector Machine (SVM) and Chaotic Neural Network. From the algorithms applied, the proposed NEAT algorithms performs better in terms of recognition rate on both databases used where the recognition rate achieved above 95.6%.

Early diagnosis and detection of breast cancer can be improved by deploying automated breast cancer applications. However, efficient algorithms have to be developed to detect texture features or morphological features or descriptor features that can possibility detect the presence of abnormalities in the breast. In this research work, image enhancement techniques, breast segmentation techniques, feature representation and classification methods have been explored and applied on mammograms and ultrasound images obtained from mini-MIAS and BCDR repositories. To predict the presence of lesions in images, Bayesian Neural Network (BNN) was adopted. This technique provides a sensitivity of 100% and is capable of extracting features from both mammograms and ultrasound images. To determine whether an image contains calcifications, which is a sign of the presence of cancer, support vector machine has been explored. The performance of the application is provided in terms of sensitivity, specificity, false positive and false negatives.

Keystroke dynamics, which distinguishes individual by its typing rhythm, is the most prevalent behavior biometrie authentication system. Neural Network is the active research area where different area has been presented. This paper present a keystroke dynamics Biometric system using chaotic neural network as the dimensional reduction and pattern recognition of the individual. Biometric scheme are being extensively used as their security qualities over the prior authentication system based on their history, that is the records were easily lost, guessed or forget. Biometric is more complex than password and is unique for each individual. In this work, the focus is made on the dwell time and flight time of the users' typing to recognize or reject an imposter. For this paper, the recognition rate obtained for the application of chaotic neural network was 99.1%.