A clustering model that examines the relationship between accessibility and vulnerability was proposed. The aim was to observe spatial accessibility variables through urban public transportation and assess the vulnerability of the population in urban areas. This model was tested in the context of the COVID-19 pandemic in the city of Cali. A hierarchical agglomerative clustering analysis was conducted on 22 communes in Cali, Colombia, at three different points in time. The Shapiro-Wilk test was used to assess the normality of the data grouped into the obtained profiles. The model identifies homogeneous groups and defines the zones of concern within the city of Cali. These zones provide insight into the risk scenario regarding accessibility-vulnerability profiles, including access to essential services such as healthcare and food for various population groups (more vulnerable or less vulnerable). The results revealed a new zoning of the city of Cali within the context of the COVID-19 pandemic risk scenario. The accessibility-vulnerability profiles were automatically identified, supporting strategic decision-making processes related to public policies, risk management, and land use management. New zoning of the city enables targeted attention to each area based on its specific characteristics, thereby addressing the complexity of the problem.
Recently, student dropout has been a major challenge for higher education institutions, since high dropout rates generate negative consequences, both academic and social, as well as economic losses for all involved. For these reasons, this study proposes a machine learning-based approach for the early prediction of dropout risk, based on academic, socioeconomic, and demographic variables. The methodology started with an exploratory data analysis, followed by a class balancing of the dataset using the SMOTE technique. Subsequently, supervised classification algorithms such as Support Vector Machines (SVM), XGBoost, Random Forest, and Balanced Random Forest Classifier were trained to build the prediction models. GridSearchCV was used to search for the best hyperparameters. The models were evaluated using metrics robust to class imbalance, such as recall, ROC-AUC curve, geometric mean (G-Mean), precision, and F1-Score. The best-performing models were Random Forest and its Balanced Random Forest Classifier variant, without hyperparameter optimization, with AUC values of 0.94 and recall values greater than 0.8 in the positive class (at-risk students). In addition, model explanation with SHAP values was used to identify and analyze the most influential risk factors (academic and socioeconomic characteristics).
High staff turnover presents a persistent challenge in the software development industry, often resulting in low productivity, loss of knowledge, and disruption of workflows. This issue becomes even more critical in settings with professional training programs for students. This paper proposes a structured self-mentoring model to mitigate the negative effects of high turnover, wherein experienced interns guide new interns, ensuring knowledge transfer and workflow stability. The approach was implemented over five internship cycles, combining the ISO/IEC 29110 standard with agile methods to support continuous development, maintain projectcontinuity, and fostergrowth in both technical and softskills.The results of project data analysis across cycles indicate improvements in productivity, quality, and participation, accompanied by a significant reduction in rework during peak performance cycles. This model demonstrates that even in high-turnover environments, structured practices can sustain quality software development while enriching students' learning experiences.
This paper presents the results of a research project focused on effort estimation in software development using supervised machine learning techniques. To structure the analysis process, the CRISP-DM methodology was adopted, given that it is recognized for its comprehensive approach and wide acceptance in data mining. The study was based on a dataset provided by the International Software Benchmarking Standards Group (ISBSG), to which rigorous cleaning, transformation, and variableselection procedureswere applied. Foureffortcategorieswere defined, and keyvariablesfortheirclassificationwere identified, including the functional size of the software, team productivity, programming language, and the implementation platform. Eight predictive models were developed using representative supervised learning algorithms: AdaBoost, Decision Trees, Random Forests, SVM, Multilayer Perceptron, KNN, Naive Bayes, and Logistic Regression. Their evaluation was carried out using metrics such as the F1-score, MCC, ROC-AUC, Gini index, accuracy, and standard deviations to assess performance and stability. The results show that tree-based models, particularly Random Forest, offer superior performance, achieving an accuracy of 80%. It is concluded that havingsystematized and high-quality data is fundamental for building reliable predictive models. As future work, the study proposes examining additional ensemble configurations, incorporating new algorithms, and using updated versions of the ISBSG repository.
Adequate daily hydration is essential for optimal health, yet one third of adults worldwide fail to meet recommended water intake. The risk is especially pronounced among adults aged 50-65, who record the lowest consumption and are at increased risk of dehydration related complications affecting digestive, respiratory, nervous systems, among others. Although several mobile interventions have been made to promote fluid intake, only a few employ evidence-based persuasive strategies or are tailored to the specific habits and motivations of middle-aged adults. This study introduces PersuDrink 50+, a persuasive mobile application designed for healthy individuals aged 50-65 to improve hydration behaviors. The app integrates goal surveys measured these constructs, while the app's built-in tracker logged daily fluid volume. Results demonstrated gains across all outcomes: knowledge scores rose from 1.4 to 3.0, perceived control from 5.75 to 8.10, attitudes from 7.36 to 9.28, and intention from 6.33 to 8.63. Correspondingly, participants increased or decreased their daily water intake to align with the WHO recommendation of >= 2 L/day. While the small sample and short test time do not enable generalizing about it, these preliminary findings suggest that PersuDrink 50+ enhances hydration knowledge and motivation in a short time frame. Future work will involve a larger randomized controlled trial and extended test time and follow up to assess the long-term effectiveness of the system.
One of the main social issues in Colombia is the consumption of illegal psychoactive substances, among which marijuana is the most consumed by individuals. Although various preventive programs have been created to reduce the consumption of these substances, few studies utilize technology to address this problem. Therefore, this study presents a machine learning model that predicts the risk of illegal psychoactive substance use in Colombia, using data from ENCSPA-2019. The model demonstrated optimal performance, achieving metrics above 98%, enabling the identification of key factors associated with consumption that influence an individual's risk level, and providing a useful tool to support public health decisions and prevention efforts.
This study explores the use of long short-term memory neural networks for the recognition of Colombian sign language. It covers both static (letters) and dynamic (word) signs. The results show that the model achieved 90% accuracy in letter recognition and 82% in words, identifying an average of 27 independent signs. In addition, different strategies for extracting spatiotemporal features using MediaPipe were analyzed, and it was found that to detect static signs only the hand control points are sufficient, and for dynamic signs the hand and posture control points are needed. However, the results did not exceed the 90% accuracy achieved in international studies, suggesting that the quality and quantity of the data set used could be improved. As future work, it is proposed that the model's performance be evaluated in real time to facilitate communication between deaf and hearing people. It is also recommended to explore more advanced deep learning architectures, such as graph convolutional networks, transformer neural networks, or combinations of convolutional neural networks with long short-term memory, as these have shown good results in dynamic sign recognition.
This work presents data modelling from Wireless Sensor Networks (WSN) applying Multiple Output Gaussian Processes (MOGP). The objective, in addition to describing the dynamics of sensed magnitudes (temperature, relative humidity, atmospheric pressure, and soil humidity), is to exploit the probabilistic learning nature of MOGPs and use the variance (uncertainty) provided by this method to reduce dimensionality in the network data, meaning the elimination of redundant data. It is compared with other machine learning models such as simple Gaussian Processes (GP), Support Vector Regressor (SVR), Neural Networks (NN), and Random Forest (RF), in terms of Root Mean Squared Error (RMSE) for modeling real data sensed by the WSN located at Universidad del Quind & iacute;o. Results demonstrate that MOGPs are highly accurate supervised learning algorithms, flexible in modeling any physical magnitude, and capable of detecting redundant data, in some cases achieving a reduction greater than 50 %.
The strategic integration of Agile methodologies, automation, and Statistical Process Control (SPC) enabled the Technology Unit of a public university to improve its software development processes, automate nineteen key performance indicators (KPIs), and establish a path toward data-driven management aligned with institutional strategy. This article presents the experience of this unit during an initiative conducted from 2021 to 2024, guided by the Stairway to Heaven model and supported by frameworks such as Scrum, BizDevOps, OKR, COBIT, and ITIL. The primary goal was to align technical processes with institutional strategy and regulatory compliance while fostering a culture of transparency, adaptability, and continuous improvement grounded in measurement. A hybrid methodology combining the Goal-Question-Metric (GQM) approach and SPC was adopted to define, monitor, and evaluate KPIs across processes including software development, incident management, and Product Owner satisfaction. Initial efforts focused on selecting value-driven indicators, followed by data collection and visualization through Power BI dashboards linked to Azure DevOps and SharePoint, which enabled real-time traceability and iterative measurement cycles. Subsequently, SPC techniques-including c-charts, X & S charts, and process capability indices (Cp and Cpk)-were applied to assess process stability and variability, identify deviations, and guide corrective actions aimed at continuous improvement. This documented experience serves as a practical guide for organizations transitioning from reactive quality control to proactive, evidence-based management tailored to regulated environments. It also provides insights into the challenges inherent in continuous improvement of software development processes.
The manufacturing sector has undergone significant transformation with the advent of Industry 4.0 and the emerging principles of Industry 5.0, integrating advanced technologies such as Cloud Computing (CC) and Big Data to optimize processes, improve operational efficiency, and enhance customer experiences. This study presents a systematic, step-by-step process for implementing a CC solution from scratch within the automotive industry, focusing on the development of an e-commerce platform to expand market reach and improve vehicle sales. The architecture was designed using Amazon Web Services (AWS) as the cloud provider, considering multiple technical and economic attributes. Industry experts evaluated the proposed solutions using the Analytical Hierarchy Process (AHP), incorporating critical factors such as performance, cost-effectiveness, and scalability to select optimal architecture. A System Dynamics (SD) simulation model was employed to analyze and compare the selected implementations; it provided quantitative insights into their long-term impact. Results demonstrate substantial differences in performance and costs across the analyzed architectures, highlighting the importance of a structured evaluation process when adopting cloud technologies. The study offers a replicable methodology for manufacturers seeking to transition to cloud-based infrastructures, contributes to a better understanding of the economic and operational implications of CC adoption. The findings underscore the relevance of cloud solutions in aligning manufacturing operations with the objectives of Industry 4.0 and 5.0.
Una de las principales problemáticas sociales que se presenta en Colombia es el consumo de sustancias psicoactivas ilegales, entre las que la marihuana es la sustancia más consumida por las personas. Si bien se han creado diversos programas preventivos para reducir el consumo de estas sustancias, son pocos los estudios que usan la tecnología para abordar esta problemática. Por tanto, este estudio presenta un modelo de machine learning que predice el riesgo de consumo de sustancias psicoactivas ilegales en Colombia, con datos de la ENCSPA-2019. El modelo demostró un rendimiento óptimo y logró métricas superiores al 98 %, lo que permite identificar factores clave asociados al consumo que influyen en el nivel de riesgo de una persona y ofrece una herramienta útil para apoyar decisiones en salud pública y prevención.
The rapid evolution of software development has heightened the need to ensure high-quality products through systematic, continuously improved processes. In response, the Software Development Center (CEDESOFT) implemented a web-based system based on the Goal-Question-Metric (GQM) method to manage key project metrics. From 2022 to 2024, the system tracked 22 projects, recording estimation accuracy compared to the previous average of 4.54%. Cost deviation was reduced from 12% to 3.5%, and post-delivery defects decreased by 92%. These results suggest that the system contributed to more precise planning, enhanced quality control, and adoption of a metrics-driven organizational culture. Limitations included scarce pre-implementation data and the need for iterative metric refinement.
Software development processes face the constant challenge of improving quality controls within the project's construction without affecting operational efficiency and meeting customer needs. DevOps offers a potential solution by enabling software development with the best practices; however, the disadvantage of frameworks such as DevOps is that they indicate that they create the practices, but not how to implement them through precise guidelines and under specific tools, leaving this step to an experimental process of trial and error, which can sometimes be costly. To address that issue, this article proposes a guide that facilitates the step-by-step adoption of five practices: version control, change requests controlled with manual code inspection, continuous integration, static code analysis, and implementing an automated pipeline for continuous integration. The methodology involves: 1) identifying software development best practices and organizing them into a step-by-step process that allows for phased implementation; 2) detailing the steps to implement each practice with specific technologies; 3) practical application; and 4) analysis and discussion of the results. The guide was presented to students, who used it to develop a short course project. Implementing the guide's practices allowed them to recognize that the information from different tools allows for quality control as the project evolves, making the process more efficient.
Immersive technologies, those that enable interaction between reality and a simulated environment, have gained relevance in higher education due to their ability to improve student motivation and learning. However, teachers face challenges when implementing them, for instance, the lack of adequate authoring tools and their growing importance in the educational context. These applications allow people without software development experience to create educational and multimedia digital content intuitively and to create content using immersive technologies. This work aimed to characterize technological, didactic, and pedagogical aspects to establish a reference framework that facilitates the development of authoring tools to create immersive content. Through a systematic review, key elements of design, architecture, technology, and functionality were identified, as well as relevant didactic and pedagogical aspects. Technological considerations such as the use of web technologies, content management, application distribution, editing interfaces, resource libraries, and content export options stand out among the findings. These results demonstrate the need and relevance of establishing a reference framework for the development of immersive applications. Consequently, a preliminary version that incorporates key design aspects and establishes solid foundations for future research and developments in the field of authoring tools with immersive technologies in higher education is proposed.
The increasing implementation of algorithmic systems in various areas of society has created a noticeable tension between algorithmic ethics and human ethics. This research examines this tension and proposes strategies to achieve a responsible balance between both ethical perspectives. The study focuses on three fundamental questions: What are the main points of conflict between algorithmic ethics and human ethics in automated decision-making processes? What approaches could harmonize these two ethical perspectives? How can human ethical principles be effectively incorporated into the design and operation of algorithmic systems? The main findings reveal that conflicts arise primarily in areas such as transparency, accountability, biases and discrimination, and the preservation of human autonomy. The research identifies key strategies to address these challenges, including the implementation of "ethics by design" frameworks, the development of ethical auditing processes, the promotion of diversity in development teams, and the advancement of digital ethics education. It is concluded that the effective integration of algorithmic ethics and human ethics requires a holistic approach that combines technical advancements, ethical reflection, ongoing education, and adaptive regulation. Only through this multifaceted effort can a technological ecosystem be created that enhances human capabilities and promotes a fairer and ethically conscious society in the digital age.
Despite being the initial and fundamental actor in specialty coffee's global production and marketing chain, coffee growers receive a disproportionately small share of the product's final value. This situation, aggravated by the adverse effects of climate change, market volatility, and increased production costs, exposes them to a precarious economic situation. There are many ways to help these producers, such as estimating production costs and selling prices, estimating climate, improving infrastructure, supporting the creation of cooperatives, and promoting fair trade. This review aims to identify several methods employed across different countries for estimating the price of specialty coffee. For this purpose, a systematic review methodology is conducted. This methodology involves identifying the need for the review, selecting and analyzing primary studies, and disseminating findings. Several studies employ mathematical models based on regression analysis, time series models, and artificial neural network-based models for coffee price estimation. In addition, it was identified that the most used evaluation metrics for the models mentioned above were R2, Akaike Information Criterion (AIC), and Average Mean Squared Error (MSE), respectively. Findings also reveal that these models often used data from online auctions, supermarkets, and the stock market, emphasizing the Cup of Excellence (CoE) auction contest. It was also identified that the main analyzed variables for the estimation were date or year, price of coffee, country of origin and destination, variety, ranking in the auction, and altitude. The study emphasizes the need for new methods and variables to estimate specialty coffee prices and their potential positive impact on the industry. These methods must adjust parameters flexibly as the models are prone to change over time.
This article describes the process, tools, and technological platforms used to implement a computer system that collects, processes, transforms, stores, and analyzes real-time traffic incidents. This system, fed by data from the Waze application accessed through the Waze for Cities program, aims to support decision-making in urban traffic management. Its implementation uses a spiral methodology that generates evolutionary deliveries, progressing towards more complete versions. The architecture, components, and data flow are detailed, allowing to get interactive and summarized views of the information, including filters, analysis, summaries, statistics, and maps. An application case is presented in the city of Bogotá, where the Secretariat of Mobility uses the real-time traffic incident monitoring system and integrates it with other information systems to identify incidents, congestion, or roadblocks, anticipating problems and planning strategies focused on optimizing mobility in the city. The system can collect and process large amounts of information, perform advanced data analysis to identify mobility patterns, anticipate critical points, and predict contingencies such as accidents or blockages. This information facilitates a more agile and accurate response by transport authorities and strengthens evidence-based decision-making, contributing to more efficient, safer, and citizen-centered traffic management.
This research addresses the optimal reactive power compensation problem in medium-voltage distribution networks by integrating Thyristor-based Switched Capacitors (TSCs) and applying the Artificial Hummingbird Algorithm (AHA). Using TSCs in distribution networks aims at reducing the annualized grid operating costs associated with energy losses while considering the costs of investment in reactive power compensators. The AHA-a bio-inspired metaheuristic optimization approach—is used to determine the optimal set of nodes and sizes for the TSCs. To estimate the expected annualized costs of energy losses, a classical power flow approach based on successive approximations is implemented, structured within a master-slave framework with the AHA. Numerical results on 33- and 69-bus systems demonstrate the effectiveness of the proposed approach compared to three other metaheuristic methods: the sine-cosine algorithm, the Chu & Beasley genetic algorithm, the particle swarm optimizer, and the black widow optimizer. All computational validations were performed using MATLAB software.
Este estudio explora el uso de redes neuronales de memoria larga a corto plazo para el reconocimiento del la lengua de señas colombiana. Abarca tanto señas estáticas (letras) como dinámicas (palabras). Los resultados muestran que el modelo alcanzó una precisión del 90 % en el reconocimiento de letras y del 82 % en palabras, y se logró identificar en promedio 27 señas independientes. Se analizaron, además, distintas estrategias de extracción de características espaciotemporales por medio de MediaPipe y se encontró que para detectar señas estáticas solo bastan los puntos de control de manos y para señas dinámicas se necesitan los puntos de control de manos y postura. Sin embargo, los resultados no superaron el 90 % de precisión alcanzado en estudios internacionales, lo que sugiere que la calidad y cantidad del conjunto de datos utilizado podría mejorarse. Como trabajo futuro, se plantea evaluar el desempeño del modelo en tiempo real, con el fin de facilitar la comunicación entre personas sordas y oyentes. También se recomienda explorar arquitecturas de aprendizaje profundo más avanzadas, como redes convolucionales de gráficos, redes neuronales transformer o combinaciones de red neuronal convolucional con memoria larga a corto plazo; estas han mostrado buenos resultados en el reconocimiento de signos dinámicos.
En este trabajo se presenta el modelamiento de datos provenientes de redes de sensores inalámbricos (WSN), aplicando procesos Gaussianos de múltiples salidas (MOGP). El objetivo, además de describir la dinámica de las magnitudes medidas (temperatura, humedad relativa, presión atmosférica y humedad de suelo), es aprovechar que los MOGP son máquinas de aprendizaje probabilísticas y usar la varianza (incertidumbre) que entrega este método para reducir la dimensionalidad en los datos de la red, es decir, eliminar datos redundantes. Se hace una comparación con otras máquinas de aprendizaje como procesos Gaussianos simples (GP), regresor basado en máquina de soporte vectorial (SVR), redes neuronales (NN), y Random Forest (RF), donde se evalúa el rendimiento en términos del RMSE, en el modelamiento de datos reales medidos por la WSN ubicada en la Universidad del Quindío. Los resultados prueban que los MOGP son algoritmos de aprendizaje supervisado de alta exactitud, flexibles para modelar cualquier magnitud física y, además, detectan datos redundantes, en algunos casos con reducción mayor al 50%.
