University of Technology, Mauritius

Ever since the introduction of Bitcoin, cryptocurrencies have attracted interest from many due to their potential for appreciation. They have also been stigmatized for their volatility, low correlation with traditional assets, and uncertain regulatory conditions. Nevertheless, many cryptocurrencies still attract the attention of major players in finance. As in the management of a portfolio of other assets, there is a need to regularly rebalance the portfolio to optimize returns across different time horizons. This study evaluates four machine learning models, Logistic Regression, Decision Tree, K-Nearest Neighbors, and Gradient Boosting, for identifying optimal cryptocurrency rebalancing decisions. Using a momentum-based approach with a 30-day forward window, the models are trained to classify assets as hold or rebalance based on future price movement. Our approach is based on feature engineering and hyperparameter tuning. Results show that tree-based models, such as Decision Tree and Gradient Boosting, demonstrate superior classification performance in identifying optimal rebalancing moments. However, the study also highlights the limitations of using historical data exclusively without referring to other external factors such as market sentiment and regulatory changes. Overall, the study makes a contribution to the field of cryptocurrency portfolio management by providing one of the first comparative evaluations of multiple ML architectures for cryptocurrency rebalancing decisions, demonstrating the potential of machine learning to improve portfolio management in highly volatile markets.

Climatic changes and anthropogenic disturbances are considered as one of the threats to continued existence of tropical marine ecosystems. These threats are complex in nature and time variant in type thus, posing a significant challenge in prediction of benthic. Moreover, to achieve improved accuracy of predictions, an advanced predictive model and optimization of its hyper parameters is required. This research aimed to predict the benthic of Flic end Flac lagoon of Republic of Mauritius. This research proposes Self Organizing Maps (SOM) clustered Deep Recurrent Neural Network (RNN), called the hybrid model (SOM Deep RNN) for high quality predictions. The approach was: i) stationarity of data was ensured, ii) feature grouping using SOM clustering was performed over the data, iii) the cluster quality was improved by optimizing its hyper parameters using methods such as random search, grid search and Bayesian hyper parameter optimization (HPO) methods. Based on validation metrics, the research identified random search as the best performing HPO method for SOM clustering, iv) clustered data were fed to Deep RNN which performed predictions, iv) prediction accuracy were improved by conducting HPO of the model, v) based on the evaluation metrics such as Root Mean square Error (RMSE), Mean Absolute Error (MAE), it was found that grid search was <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9.45\%}$</tex> (MAE), <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{6.56\%}$</tex> (RMSE) more accurate than random search and 19.81% (MAE), <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{14.37\%}$</tex> (RMSE) more accurate than bayesian search. Thus, the research determined that grid search is the best HPO for Deep RNN prediction model for the marine data under consideration.

Wasserstein Generative Adversarial Networks (WGANs) have gained significant attention due to their theoretical foundations and effectiveness in generative modeling. However, training stability remains a major challenge, typically addressed through fixed gradient penalty (GP) techniques. In this paper, we propose an Adaptive Gradient Penalty (AGP) framework that employs a Proportional–Integral (PI) controller to adjust the gradient penalty coefficient λt based on real-time training feedback. We provide a comprehensive theoretical analysis, including convergence guarantees, stability conditions, and optimal parameter selection. Experimental validation on MNIST and CIFAR-10 datasets demonstrates that AGP achieves an 11.4% improvement in FID scores on CIFAR-10 while maintaining comparable performance on MNIST. The adaptive mechanism automatically evolves penalty coefficients from 10.0 to 21.29 for CIFAR-10, appropriately responding to dataset complexity, and achieves superior gradient norm control with only 7.9% deviation from the target value compared to 18.3% for standard WGAN-GP. This work represents the first comprehensive investigation of adaptive gradient penalty mechanisms for WGANs, providing both theoretical foundations and empirical evidence for their advantages in achieving robust and efficient adversarial training.

This paper proposes a family of first order bivariate integer-valued autoregressive (BINAR(1)) with Poisson Lindley innovations (BINAR(1)PL). The model parameters are estimated using the conditional maximum likelihood (CML) estimation approach. The proposed models are applied on some real life crime data.

The purpose of this research is to explore the application of lean techniques as a means to reduce waste in order to promote sustainability within textile firms in Mauritius. This study will present some lean techniques, tools that will reduce energy, environmental wastage, raw materials utilization and improve leanness and agility. The results revealed that most of the respondent disagreed that their company was on the right track towards sustainability. A clear correlation between the absence of Lean practices and the sustainability issues the Textile industry of Mauritius is facing was found. Moreover, the study showed that the industry was facing issues like high rework and reject rate, poor people retention, high absenteeism and high energy consumption. Finally, it was deduced that all of these issues have been linked to the lack of these lean tools and techniques and ultimately to sustainability. As such reducing the 7 forms of waste can ultimately improve sustainability, thus completing all of the 3 main objectives of the study.