Rafael Rodríguez-Puentes

This study investigates the role of reduct-based attribute subsets (derived from rough set theory) with respect to supervised classification performance. Considering twenty-one datasets, two representations were assessed: minimum-length reducts containing the smallest attribute sets to retain consistency, and maximum-length reducts comprising the largest irredundant subsets. A Friedman–Nemenyi test was used to assess eleven classifiers with five quality measures. For the majority of classifiers, both types of reducts systematically degrade the predictive performance when compared with using all attributes. JRIP, Logistic Regression, and MLP are consistent and have the same performance except for PRC area, where maximum-length reducts reduce performance. For SMO, minimum-length reducts perform better than maximum-length reducts. These observations show that the efficacy of reduct-based feature selection largely depends on the characteristics of classifiers; it can provide significant dimensionality reduction without loss of accuracy for some models. Spanish-language metadata / Metadatos en españolTítulo en español:¿Importa el tamaño? La influencia del tamaño del reducto en el rendimiento de los clasificadores supervisados Resumen:Este estudio investiga el papel de los subconjuntos de atributos basados en reductos, derivados de la teoría de conjuntos aproximados (rough set theory), en relación con el rendimiento de la clasificación supervisada. Considerando veintiún conjuntos de datos, se evaluaron dos representaciones: reductos de longitud mínima, que contienen los conjuntos de atributos más pequeños necesarios para conservar la consistencia, y reductos de longitud máxima, que comprenden los subconjuntos irredundantes de mayor tamaño. Se utilizó una prueba de Friedman–Nemenyi para evaluar once clasificadores mediante cinco medidas de calidad. Para la mayoría de los clasificadores, ambos tipos de reductos reducen sistemáticamente el rendimiento predictivo en comparación con el uso de todos los atributos. JRIP, Regresión Logística y MLP presentan un comportamiento consistente y muestran el mismo rendimiento, excepto en el área PRC, donde los reductos de longitud máxima reducen el rendimiento. En el caso de SMO, los reductos de longitud mínima presentan un mejor rendimiento que los reductos de longitud máxima. Estas observaciones muestran que la eficacia de la selección de características basada en reductos depende en gran medida de las características de los clasificadores; para algunos modelos, puede proporcionar una reducción significativa de la dimensionalidad sin pérdida de exactitud. Palabras Claves:reducción de dimensionalidad, reducto, clasificadores, longitud del reducto Smart citations: SciteAI. Dimensions.Open Alex.

To address the complexity of hypergraph-related problems, several authors have proposed alternative structures for representing hypergraphs. Although these new representations may improve the understanding of such problems, their associated computational complexity often remains NP-hard. In addition, some authors have developed algorithms tailored to specific types of hypergraphs. In this study, we introduce a novel representation of simple, undirected hypergraphs using the concept of a reduced graph. This representation preserves the high-level relationships of the original hypergraph while enabling the design of efficient algorithms for specific hypergraph classes. We demonstrate that this structure is particularly useful for computing minimal transversals of minimum cardinality as well as minimal transversals of maximum cardinality. Furthermore, we present a simple algorithm to compute the latter for irredundant hypergraphs.

Learning analytics has become an essential component of evidence-based decision-making in higher education, yet many predictive models continue to face critical limitations, including weak generalizability, narrow contextual alignment, and limited integration of established student retention theories. This study advances the field by developing a theoretically grounded and ensemble learning model to predict student retention at the International University of Management (IUM) in Namibia. Using a large-scale dataset of 11,090 anonymised student records (2021–2023) comprising academic, behavioural, demographic, and engagement variables, the study applies a suite of machine learning techniques including Decision Tree, Random Forest, Support Vector Machine, Artificial Neural Network, and Gradient Boosting and an optimized stacking ensemble architecture. The dataset was divided into 80% training and 20% testing subsets, and the models were evaluated using 10-fold cross-validation to ensure stability and generalizability. Predictive performance was assessed using accuracy, precision, recall, and F1-score, with the final ensemble (DT-ANN-RF-SVM) achieving 94.6% accuracy, outperforming all individual models. The study makes three key contributions: (1) it operationalizes constructs from Tinto’s Student Integration Theory, Bean and Metzner’s Attrition Model, and Astin’s Theory of Student Involvement into measurable learning analytics variables; (2) it demonstrates how ensemble learning can overcome methodological limitations prevalent in prior work; and (3) it offers a reproducible institutional framework for applying analytics-driven retention strategies. The results highlight the value of theoretically informed machine learning models in identifying at-risk students early and enabling tailored support interventions. The findings provide actionable insights for institutional policy, pedagogical design, and future research on adaptive learning analytics in higher education.

Financial fraud remains a persistent challenge across various domains, particularly in public-sector financial operations, threatening the integrity and transparency of financial statements while eroding public trust. This highlights the need for continued advancement of fraud detection mechanisms to keep up with the ever-evolving fraud tactics. ML algorithms have proven to be one of the most successful methods for analysing large financial datasets to detect fraudulent patterns. This paper reviews the application of ML to detect fraud in financial transactions using ML-based algorithms, namely K-means, Support Vector Machine, Decision Trees, Naive Bayes, and Deep Learning, in fraud detection, analysing their use cases and effectiveness as reported in the literature. Additionally, the study experimentally compares the performance of a Convolutional Neural Network (CNN) model against a Logistic Regression model, with the CNN achieving an impressive 90% accuracy, outperforming Logistic Regression in fraud detection. The paper further investigates the financial features and indicators most relevant to fraud detection and explores the challenges and opportunities posed by large volumes of financial transactions. By addressing these areas, the study aims to provide insights into enhancing fraud detection mechanisms and strengthening the security and integrity of financial transactions in today's digital ecosystem, including government institutions.

In the context of the information age, a land administration system must be technologically driven to manage land information and data transparently. This ensures the registration and protection of land rights for people. In this study, we present a Blockchain Land Registration system designed as a tool for enhancing land administration in South Saharan Africa (SSA). Drawing inspiration from Namibia, we have developed a user interface comprising a homepage/landing page, a users’ registration form, a login form that incorporates MetaMask authentication prompts, and an authenticated dashboard for landowners and purchasers. Design Science was employed as the methodology for this proposal. Being technical design research for solving a land administration problem (that of inefficient land registration), the technical solution identified involves system design, the development of blockchain integration and testing, and development aspects. Based on this approach, blockchain was conceptualised as an “artefact” that could be investigated as a technical solution to address the challenges posed by inefficient land registration. This study provides a comprehensive roadmap for the conceptualisation, development, validation, and deployment of a blockchain-based land titles registry suitable for SSA countries. It also explores a discussion on the practical and policy implications of blockchain in land administration in SSA countries.

Pest control has been an important undertaking in maximising yields from crop farming. Manual methods of combating pests have shown limited success, prompting the need for more modern, innovative approaches. This paper presents the use of AI and IoT for pest control. The developed prototype employs AI to analyse images captured by a camera. The analysis focuses on detecting and identifying pests, which then triggers the irrigation process for the specific pesticide required. The prototype has been developed using the Design Science Research Methodology. The prototype was implemented on a Raspberry Pi with a camera and a water pump. The camera continuously captures images, which are processed by a Machine Learning model tasked with detecting and identifying pests. Once a pest is identified, the water pump is then activated to spray the appropriate pesticide based on the identified pest. The study's results demonstrate the potential for controlling pests using AI and IoT, as the prototype was able to efficiently identify pests and trigger the spraying of the pesticide. Future work will examine the full implementation of a prototype in a crop field, with additional functionality such as weather monitoring and predicting future pest infestations based on historical data.

The use of Geographic Information Systems has increased considerably since the eighties and nineties. As one of their most demanding applications we can mention shortest paths search. Several studies about shortest path search show the feasibility of using graphs for this purpose. Dijkstra’s algorithm is one of the classic shortest path search algorithms. This algorithm is not well suited for shortest path search in large graphs. This is the reason why various modifications to Dijkstra’s algorithm have been proposed by several authors using heuristics to reduce the run time of shortest path search. One of the most used heuristic algorithms is the A* algorithm, the main goal is to reduce the run time by reducing the search space. This article proposes a modification of Dijkstra’s shortest path search algorithm in reduced graphs. It shows that the cost of the path found in this work, is equal to the cost of the path found using Dijkstra’s algorithm in the original graph. The results of finding the shortest path, applying the proposed algorithm, Dijkstra’s algorithm and A* algorithm, are compared. This comparison shows that, by applying the approach proposed, it is possible to obtain the optimal path in a similar or even in less time than when using heuristic algorithms.