University of Technology, Mauritius

Data envelopment analysis (DEA) is a method for identifying the best practices among peer decision-making units (DMUs). Early DEA models were suited for a production possibility set of positive inputs and outputs. In real setups, some inputs and outputs may be negative. Several past studies have provided methods of dealing with negative data in DEA. Furthermore, the premise of early DEA models was that DMUs are homogeneous, though DMUs may exhibit a network structure with negative input and output values in practice. In this study, we presented semi-oriented radial measure (SORM) network DEA models to assess the system as a whole as well as the first and second Stages. The proposed method was compared with a network bi-directional SORM (DSORM) model. The targets obtained from the SORM network model revealed an improvement from the observed values, whereas some of the Stage 2 targets from the DSORM model revealed reductions from the observed values. An integrated SORM model was used to analyse the overall system at once; a higher discrimination of the DMUs was observed in the model. The models were applied to data from 13 insurance companies in Mauritius. The premium collection Stage (Stage 1) was observed to be the dominant Stage of the insurance companies.

Credit scoring is an essential tool used by financial institutions to estimate the creditworthiness of individuals and businesses. It represents a numerical evaluation of a borrower's likelihood to repay a loan and categorizes applicants into 'Good' and 'Bad' classes. Institutions rely on these scores for lending decisions, portfolio risk management, and loan conditions such as rates, limits, and repayment schedules. From a customer's perspective, a higher score provides access to better financial products, while a lower score restricts opportunities and increases costs. Twelve machine learning models were tuned and evaluated for credit scoring in this study. Moreover, detailed steps are provided for data preprocessing to allow for reproducibility and SHAP metrics were used to provide transparency to stakeholders. Models were implemented in Python, with CatBoost performing best with an F1-score of 0.88, followed by LightGBM with 0.87.

New wind energy projects in Small Island Developing States (SIDS) present a significant opportunity to reduce dependence on costly imported diesel fuels and advance energy independence. Key benefits include reducing reliance on volatile fossil fuel imports, stabilizing local economies, lowering electricity generation costs over the long term, and mitigating greenhouse gas emissions. Despite these advantages, the development of wind energy projects in SIDS is constrained by several significant challenges. For SIDS, which depend heavily on imported refined petroleum products, these challenges often translate into imported inflation, volatile electricity prices, and persistent fiscal burdens. This paper examines the opportunities and challenges associated with the deployment of new wind energy projects in selected SIDS across the Indian and Pacific Oceans. Special attention is given to the integration of Battery Energy Storage Systems (BESS) with wind power generation as a strategy to address the inherent intermittency of wind resources and improve the reliability of electricity supply.

Achieving high-fidelity 3D reconstructions of clothed humans from a single image is pivotal for applications in virtual reality, gaming, and the fashion industry. However, the challenge of accurately reconstructing subjects in loose clothing and complex poses has yet to be fully addressed. To bridge this gap, we propose a novel hybrid approach that combines decoupled side-view features and rebalanced parametric body model prior to guide detailed 3D human reconstruction. This approach can handle loose clothing and unusual poses simultaneously. Specifically, guided by the SMPL-X-based parametric human body prior, we leverage a Transformer-based framework to effectively decouple side-view features from input images. This process significantly enhances the accuracy of implicit-function-based reconstruction for complex poses, enabling a more precise representation of human body postures. Furthermore, to better adapt to diverse clothing types and avoid overfitting to the training data mainly consisting of tight clothing, we introduce a rebalancing coefficient within a positional embedding-based strategy. This coefficient adjusts the model’s reliance on the parametric body prior, enhancing the ability to capture details of loose clothing. Consequently, the model can generate more reliable Signed Distance Function (SDF) values, which are essential for creating high-fidelity 3D clothed human bodies. Extensive experiments demonstrate superior performance in detailed representations for loose clothing and maintaining robust reconstruction of complex poses.

Machine learning (ML)-based intrusion detection systems (IDS) achieve high detection accuracy but suffer from a lack of interpretability that limits their adoption in operational security environments. Explainable Artificial Intelligence (XAI) methods like SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) have emerged as prominent tools to address this transparency gap. However, few studies have systematically compared these methods under unified and reproducible conditions within the IDS domain. This paper presents an empirical comparison of SHAP and LIME applied to a Random Forest classifier trained on the CICIDS2017 benchmark dataset. After preprocessing, the dataset retains 57 features across 12 attack classes. The classifier achieves an overall accuracy of 0.9972 and a macro F1-score of 0.89. SHAP and LIME are evaluated on four criteria: computational overhead, global feature importance agreement, Spearman rank correlation, and explanation stability. The results show that LIME completes in 352.33 seconds compared to 525.63 seconds for SHAP, despite SHAP using its tree-optimised TreeExplainer variant. Both methods identify Destination Port as the most influential feature; they share 12 of the top 20 features and the Spearman rank correlation is 0.5298. Stability analysis reveals that SHAP achieves perfect reproducibility (1.0) while LIME remains near-perfectly stable (0.9897), which challenges the commonly held assumption of LIME instability. These findings provide empirical guidance for the selection of explainability methods in practical cybersecurity applications.