Advances in Computing

Plant diseases have effects on the growth and production of the plant. Plant diseases can be figured out by using digital image processing, and nowadays, Deep learning has made a lot of progress in digital image processing to identify the disease efficiently. This paper finds the plant diseases using EfficientNet by focusing on three data steps: pre-processing, model selection, and detection network, using a canny edge detection algorithm. The model is trained and tested on plant disease data set. The model provides 97.2% accuracy in detecting the disease than existing CNN-based models.

The rapid and uncontrolled population growth and the development of various industrial sectors have accelerated the rate of changes in land use and land cover (LULC). The quantitative assessment of changes in LULC plays a fundamental role in understanding and managing these changes. Therefore, it is necessary to examine the accuracy of different algorithms for LULC mapping. We compared the performance of three deep learning architectures (PSPNet, U-Net, and U-Net++) with four different backbones, including ResNet-18, ResNet-34, ResNet-50, and ResNext50_32x4d pre-trained on ImageNet. Besides, we compared the model’s performance using the same scene but using: 1) a single date, 2) a time series, and 3) data augmentation. For this, we used Sentinel 2 images captured on Antioquia-Colombia and four main categories of the Corine Land Cover as ground truth. The mean Intersection-Over-Union (mIoU) metric and pixel accuracy was used like evaluation metrics. All models showed an increase in performance with data augmentation. The best models were U-Net with ResNet-50 encoder and U-Net with Resnext50-32x4d, with pixel accuracies of 88.6% and 89.2%, respectively, and mIoU 74.6% and 74.8%. Both models had similar computing times (244.07 min and 248.06 min). PSPNet was the lowest-performing architecture, with pixel accuracy between 83.2% and 84.1% and mIoU between 63.3% and 64.6%. In summary, our results show that semantic segmentation models are suitable for classifying the LC of optical images and provide benchmark accuracy for evaluating the integration of new techniques and sensors.

Land use and land cover classification (LULC) is a fundamental input for ecological and socioeconomic models worldwide, generating a large volume of data from space-based platforms, mainly optical technologies. However, these can be affected by atmospheric conditions. Colombia has a high percentage of cloud cover due to its geographical location, which makes it challenging to map LULC changes. Studies have emerged that evaluate the integration of optical and radar images with algorithms that allow for good results despite the information gaps that affect these processes. Therefore, this work compares three supervised machine learning approaches, Support Vector Machines, Random Forest, and XGBoost, to classify land use and land cover from multispectral and radar images, contemplating four scenarios for data fusion. Optical, optical + SAR, optical + SAR ascending, and optical + SAR descending. The result for the Random Forest model using optical + ascending SAR has the best accuracy (76.02%), followed by Random Forest with optical + descending SAR data (75.97%) and with little difference for Random Forest using optical data (75.83%). In future studies, it is of great interest to explore feature extraction on both data sets to improve LULC representation and classification.

Computer rooms are essential spaces in higher education institutions, they are used for classes, laboratories, and practice sessions related to the various academic programs. In order for students to be able to use computer room equipment without any inconvenience, it must be in perfect working order. Nevertheless, it is very common for computer room equipment to experience malfunctions as a result of excessive or improper use. Therefore, it is essential that these malfunctions are identified and communicated promptly so that they can be resolved by those responsible for equipment maintenance. To address this need, SisReport is proposed as a platform that allows students affected by device failures to make a direct report to the maintenance area. The proposed system will be evaluated from the point of view of user experience. The evaluation of the first version of SisReport involved students, teachers, and the Information and Communications Technology (ICT) team of the Institución Universitaria Colegio Mayor del Cauca. The findings of the evaluation showed that the proposed system is effective for notifying equipment failures in computer rooms compared to the current manual way. The system described in this article can be used by other institutions of higher education to facilitate students’ reports of failure of computer equipment in the computer rooms.

Differential cryptanalysis has proven to be a powerful tool to identify weaknesses in symmetric-key cryptographic systems such as block ciphers. Recent advances have shown that machine learning methods are able to produce very strong distinguishers for certain cryptographic systems. This has generated a large interest in the topic of machine learning for differential cryptanalysis as evidenced by a growing body of work in the last few years. In this paper we aim to provide a guide to the current state of the art in this topic in the hope that a unified view can better highlight the challenges and opportunities for researchers joining the field.

In Colombia there is a problem related to eating habits that has its origin, mainly, in two causes: the lack of budget that allows access to a wider variety of food, and the lack of awareness among the population about their nutritional needs. To tackle this issue, a solution has been proposed using a Constraint Programming (CP) approach enhanced with Machine Learning (ML) for a version of the Diet Problem (DP).A CP model was developed to find a shopping list that meets a family’s nutritional needs while minimizing costs; and a synthetic dataset was created to test the model, which was run multiple times to collect results. Since DP is an NP-complete problem and computational time to find optimal solutions varies from one solver to another, a ML classifier was used to choose a solver that best performs in small cap time limits based on instance features (i.e., selection from an Algorithm Portfolio). After carrying out an extensive evaluation of the CP model, including our approach that implements a Classifier for algorithm selection, the model correctly selects the best solver over 68.07% of the time, for a sample of 1378 instances.By analyzing the performance of different solvers on a set of instances, it can be predicted which solver is likely to achieve the best results on new instances. This approach could be extended to tuning solver parameters, which would further improve their efficiency and effectiveness. (The dataset used for the creation of this paper is available on: https://github.com/Git-Fanfo/dataset_CCC )

ChatGPT is an advanced large language model (LLM) capable of generating code to solve specific problems when presented with carefully designed prompts, among other capabilities. The existence of ChatGPT raises signifi-cant questions regarding teaching practices and evaluations within the dis-cipline. If ChatGPT can effectively solve exercises assigned to students, it prompts a reevaluation of the skills and knowledge that we teach and eval-uate. The objective of this paper is to assess the proficiency of ChatGPT in solving exercises commonly encountered in a CS1 course. This serves as an initial step in exploring the implications of ChatGPT for computer science education. By examining ChatGPT’s performance and comparing it with real students, we aim to gain insights into its capabilities and limitations. Our evaluation encompasses a comprehensive examination of 125 problems specifically designed for CS1-level learners. The experiment revealed that ChatGPT successfully solved approximately 60% of the provided prob-lems. Subsequently, we conducted a detailed analysis of the characteristics of the problems that ChatGPT could not solve, aiming to gain a deeper un-derstanding of the nuances that make them challenging for LLMs. This study contributes to the ongoing discourse surrounding the integration of AI-based tools, such as ChatGPT, in computer science education, and high-lights the need for a reevaluation of educational objectives and methods employed in traditional educational institutions.

In software development, following an architecture is extremely essential for any project. Clean Architecture, since 2017, has become popular among the Native Android development community. It helps to improve the efficiency of the development process by establishing a clear separation of concerns, achieving modular, scalable, and maintainable code. Its advantages and how it can improve the efficiency of development projects in the Android ecosystem are shown. It examined the challenges faced by Android developers and how Clean Architecture moving from its original version to a shorter version that revolves around native Android development can solve them. The main objective is to introduce the adaptation of the original Clean Architecture model to the current state of native Android development, all oriented towards an application called “InstaFlix”, by creating a shorter and more coupled format. The developer will follow best practices and promote the use of presentation patterns such as Model-View-ViewModel (MVVM) or Model View Presenter (MVP), as well as encourage dependency injection. In short, it makes it easier for many developers to work simultaneously on different parts of the system. This separation also generated a noticeable improvement in the code base, as changes can be made to specific components without affecting the rest of the system; focusing on modularity and maintainability, making it clear that it is valuable for any Android developer who wants to create high-quality software.

This paper evaluates techniques for improving the use of cloud computing resources through autoscaling. Autoscaling, also referred to as auto-scaling or automatic scaling, is a cloud computing technique for dynamically allocating computational resources. Autoscaling can be reactive (responding to resource needs as they arise) or proactive (anticipating future demands). Our study proposes the use of AI-based models to predict the creation of new computational entities under varying load conditions. The proposed methodology included data cleaning, correlation analysis to select relevant features, and the evaluation of several supervised and unsupervised machine learning models. The results shown that machine learning techniques can be used to anticipate and optimize the capacity of computing systems.

Industry 4.0 enabling technologies are impacting several sectors of the economy by automating organizational processes. As a result, the demand for professionals with the necessary skills to face the challenges imposed by the new industrial trends has increased. This represents a challenge for technological and university institutions as they are at the forefront of the training and educational transformation processes. While there are various governmental strategies to provide training in technological topics, the most active stakeholders in implementing such strategies are higher education institutions, where the shortage of students in the foundational careers that support I4.0 technologies is particularly noticeable. For this reason, universities, as well as elementary and high schools, are consolidating strategies that could enhance the motivation of young people to choose technology-based careers.This paper presents an educational experience in Education 4.0 focused on enabling technologies for I4.0 oriented to high school students and led by The University of Medellín. Implementing this educational experience enabled the participants to devise solutions to problems real-world challenges using the Internet of Things. Among the benefits of this strategy, it does not require prior training in computational thinking or electronics fundamentals. Additionally, it integrates didactic strategies such as co-creation, gamification, and project-based learning, while allowing students to build a basic prototype within a relatively short period.

The communication gap between deaf and hearing people remains a significant issue today, as the language used by both parts creates barriers affecting multiple aspects on people’s lifes. To address this, we propose a system for translation of text and audio into Colombian Sign Language using an accessible and user-friendly mobile application to facilitate communication between both parties. For this prototype, 76 phrases and words that represent signs were selected for the system. Pre-recorded videos of a sign language interpreter were employed to perform translations based on the previously established vocabulary. The service works via video concatenation based on the client’s text or audio request, using a Whisper AI model for audio transcription. In the context of an undergraduate thesis project, a mobile application prototype that translate text and audio into Colombian Sign Language was successfully developed. This project contributes to bridging the communication gap between deaf individuals and the hearing community.

Nowadays, several video movement classification methodologies are based on reading and processing each frame using image classification algorithms. However, it is rare to find approaches using angle distribution over time. This paper proposes video movement classification based on the exercise states calculated from each frame’s angles. Different video classification approaches and their respective variables and models were analyzed to achieve this, using unstructured data: images. Besides, structure data as angles from critical joints Armpits, legs, elbows, hips, and torso inclination were calculated directly from workout videos, allowing the implementation of classification models such as the KNN and Decision Trees. The result shows these techniques can achieve similar accuracy, close to 95%, concerning Neural Networks algorithms, the primary model used in the previously mentioned approaches. Finally, it was possible to conclude that using structured data for movement classification models allows for lower performance costs and computing resources than using unstructured data without compromising the quality of the model.

Self-driving robots have to fulfill many different operations, as coordinating the motors’ traction, camera movement, or actuator arms mechanics, as well as more high-level operations like driving to different places. Autonomous navigation is of utmost importance for exploration robots, which must drive around exploring areas with unknown terrain conditions, as for example is the case of mars rovers and other space exploration vehicles. Given that the environment is unknown, planning a specific route and driving plan is challenging or even inappropriate due to blocking obstacles in the terrain. To overcome such problems we propose an adaptable plan for driving robots in different situations. Our solutions mixes both global and dynamic planning algorithms to take advantage of available information, if it exist beforehand, and to overcome unknown obstacles if they appear, while still moving towards the goal. In particular, we apply our algorithm to the movement of robots between posts in environments with partial information, as it is the case of space mission competitions. We evaluate our solution in a simulated environment taking into account the effectiveness in fulfilling a mission in the shortest time, using the shortest possible path. Our results show that of the A* algorithm with diagonals in combination with the ABEO algorithm offer the best combination reaching the goal in most cases, in optimal (planning + execution) time.

This paper considers the integration and application of a Cyber-Physical System (CPS) and a digital twin to control a three-degree-of-freedom (3DoF) robotic manipulator. Here, framed in Industry 4.0, we consider robots as interconnected components within a broader network. Supported by current literature, we contribute to advancing interlinked systems that mirror the physical dynamics of equipment and facilitate their remote visualization–a cornerstone in the architecture of Internet of Things (IoT) robotics. Our strategy is rooted in three core stages: modeling, simulation, and implementation, and aims to seamlessly integrate the constituent elements of a robotic agent within an Internet of Robotic Things (IoRT) environment. At this nascent stage, the system has undergone testing at the prototype level, with ambitions to scale it for deployment in industrial settings. Preliminary results demonstrate the efficacy of the system in simulating and controlling the robotic manipulator, highlighting the potential of this integrated approach in practical applications. Our findings are pivotal to these concepts’ evolution and roll-out, bolstering understanding of the nexus between CPS, digital twins, and robotics within Industry 4.0.

In the context of Visual Programing for Product Design, the endowment of the Designer with programing tools to boost productivity is central. However, Product (and Architectural) Design are usually taught without programing courses. This manuscript reports the results of Lattice DesignVisual Programming by a Product Designer with no previous exposure to programing but provided with the intuitive concepts of Pre-, Post-condition and Invariant logical first-order predicates for imperative programing. The scenario of application is the population of 3D domains (i.e. solid models) with lattice individuals of the type zero-curvature Truss (colloquially called 1.5D and 2.5D) structural elements. Result show that, although Pre-, Post-condition and Invariant are devised for imperative programing, they provide a solid and successful structure for visual programming (e.g. Grasshopper) for Designers with no mathematical or programming background. Regarding the specific Additive Manufacturing scope, the manuscript depicts the population of the target domain with lattice individuals which, in this case, undergo a rigid transformation before docked in the target domain. The lattice design presented allows for the grading of the lattice geometry. Future work addresses the programing of non-rigid transformations (non-affine, non-conformal, etc.) which dock the lattice individual into the target solid domain. Regarding the endowment of non-programmer Product Designer with visual programing and pre-, post- and invariant conditions, the performance results are very positive. However, as with any work team, experts must be recruited to help with highly specialized topics (e.g. computational mechanics, differential geometry, discrete mathematics, etc.).

One of the main challenges of building commercial solutions with Supervised Deep Learning is the acquisition of large custom-labeled datasets. These large datasets usually fit neither commercial industries’ production times nor budgets. The case study presents how to use Open Data with different features, distributions, and incomplete labels for training a tailored Deep Learning multi-label model for identifying waste materials, type of packaging, and product brand. We propose an architecture with a CBAM attention module, and a focal loss, for integrating multiple labels with incomplete data and unknown labels, and a novel training pipeline for exploiting specific target-domain features that allows training with multiple source domains. As a result, the proposed approach reached an average F1-macro-score of 86% trained only with 13% tailored data, which is 15% higher than a traditional approach. In conclusion, using pre-trained models and highly available labeled datasets reduces model development costs. However, it is still required to have target data that allows the model to learn specific target domain features.

Mathematics, specifically in the field of Algebra, becomes complicated for students to understand and this is reflected in their low academic grades. Additionally, in the results of reports of the international program for the evaluation of students in which Colombia is located with a lower score than the OECD average. This problem generates stress and anxiety, which can affect their ability to concentrate and retain information. The social and affective environment can also be an important factor in the learning of algebra. The methodology used for the selection of documents was systematic mapping. A total of 138 documents were found and, applying inclusion and exclusion criteria, 40 of these were selected. The most relevant current trends were found, which are the use of educational software and strategies such as gamification. It is important to find effective didactic tools that have the capacity to teach algebra in an effective way to students.

Manual measurements and visual inspection of trees are common practices among farmers, which incur labor costs and time-consuming operations to obtain information about the state of their crops at a specific moment. Considering that an approximately 1-hectare (ha) plot of land can have up to 1100 planted trees [1], this becomes a challenging task, and human error in such cases tends to be high. To address these issues, the emphasis is placed on the use of Convolutional Neural Networks (CNNs); however, CNNs alone are not robust enough to detect complex features in any given problem. Therefore, this article proposes a model that supports agricultural activities in organizing their tasks. The main procedure of the model is the classification of fruit trees (mango, citrus, and banana) using aerial images captured by a drone (UAV) in the Colombian context. The technique employed in this procedure is known as Mask R-CNN, which enables automatic segmentation of fruit trees.

This article introduces an extension for change point detection based on the Minimum Description Length (MDL) methodology. Unlike traditional approaches, this proposal accommodates observations that are not necessarily independent or identically distributed. Specifically, we consider a scenario where the counting process comprises observations from a Non-homogeneous Poisson process (NHPP) with a potentially non-linear time-dependent rate. The analysis can be applied to the counts for events such as the number of times that an environmental variable exceeded a threshold. The change point identification allows extracting relevant information on the trends for the observations within each segment and the events that may trigger the changes. The proposed MDL framework allows us to estimate the number and location of change points and incorporates a penalization mechanism to mitigate bias towards single regimen models. The methodology addressed the problem as a bilevel optimization problem. The first problem involves optimizing the parameters of NHPP given the change points and has continuous nature. The second one consists of optimizing the change points assignation from all possible options and is combinatorial. Due to the complexity of this parametric space, we use a genetic algorithm associated with a generational spread metric to ensure minimal change between iterations. We introduce a statistical hypothesis t-test as a stopping criterion. Experimental results using synthetic data demonstrate that the proposed method offers more precise estimates for both the number and localization of change points compared to more traditional approaches.

In this paper it is presented an exploration of different strategies to generate solutions in a genetic algorithm for the detection of multiple change-points in univariate time series. The purpose is to find which combination of these is the optimal one while modelling times where there is an exceedance from a given threshold through Non Homogeneous Poisson Processes. Likewise, elements from information theory are taken to define a parsimonious model such that the explained phenomenon has a low memory usage and an optimal quantity of parameters which are estimated through a Bayesian approach. These elements define the objective function. Thus and after evaluating different operators it is found that the optimal strategy to generate and to combine new solutions is through a random keys initialization, selection of the parents through the ranks and Boltzmann tournament method or through a roulette strategy and using a fixed low mutation rate such that the diversity component is supplied through a neighborhood exploration while keeping the fitness of the solutions close to the real value.

The classification of avocado maturity is a challenging task due to the subtle changes in color and texture that occur during ripening and before that. Hyperspectral imaging is a promising technique for this task, as it can provide a more detailed analysis of the fruit’s spectral signature compared with multi-spectral data. However, the acquisition of hyperspectral data can be time-consuming and expensive. In this study, we propose a method for generating synthetic hyperspectral data of avocados. The synthetic data is generated using a generative adversarial network (GAN), which is trained on a small dataset of real hyperspectral im-ages. The generated data is then used to train a neural network for avocado maturity classification. The results show that the neural network trained on synthetic data achieves comparable accuracy to a neural network trained on real data. Additionally, synthetic data is much cheaper and faster to generate than get real data. This makes it a promising alternative for avocado maturity classification.

Climate change has caused strong variations in agroclimatic parameters such as precipitation, temperature, and relative humidity, accelerating the phytosanitary conditions associated with agricultural crops, mainly in insect pests, since these generate an alteration in their life cycle and an increase in their population. This causes significant economic damage to important crops such as the Hass avocado, which has had a growing development and demand in national and international markets, which has generated significant income for small and medium-sized farmers and exporters of this fruit in the country. To mitigate the impacts of climate change on agricultural production, it is possible to implement digital agriculture technologies. These technologies allow estimating the incidence of climate variations on crops through the monitoring of agroclimatic and phytosanitary variables that affect fruit growth. Therefore, a variable dispersion model with fuzzy characterization is proposed that seeks to establish a correlation between rainfall and the aggregate distribution of losses in the Hass avocado crop. In order to analyze and validate the proposed model, the random variables related to phytosanitary risk were taken and characterized. Subsequently, the frequency and severity random variables were modeled as linguistic random variables using fuzzy logic concepts. The results indicate that rainfall is the key variable to correlate in the search for an index insurance model based on agricultural risk, as well as in the characterization of qualitative and quantitative risks, promoting the improvement of financial and environmental sustainability by reducing agricultural losses through better crop management.

The Raft consensus algorithm is widely recognized for its practicality and comprehensibility in achieving consensus within distributed systems. This paper presents a comprehensive exploration of Raft, making clear key concepts and verifying critical properties. We delve into the fundamental components of Raft, encompassing leader election, log replication, and safety guarantees. Detailed explanations are shown in order to illustrate the interactions between actors during commit phases, leader selection, and other significant stages. The Athena proof system is employed to verify essential properties such as leader completeness, log consistency, and fault tolerance, ensuring the algorithm’s resilience in the face of failures. Drawing upon the Athena programming language’s actor model implementation, we simulate and validate the behavior of Raft, providing practical insights into its functionality.

In precision agriculture, the accurate and timely plant disease identification is crucial. However, the lack of accuracy in current detection systems hampers reducing pesticide and fertilizer usage, causing significant productivity losses. The desired level of precision has not been achieved yet, hindering timely intervention and mitigation strategies. This research presents a novel approach that integrates a Lagrangian Gaussian Puff Dispersion Model (LGPTM) for assessing plant health, with Gaussian bell curve visualization, a tool for visualizing the distribution patterns of these indices in the field of precision agriculture. This integration ameliorates disease detection and monitoring in agricultural contexts, thereby improving disease management practices and enhancing crop health and productivity. The methodology leverages widely adopted libraries to process multispectral images and calculates vegetation index values based on the Normalized Difference Vegetation Index (NDVI). Additionally, the modeling approach employed modular programming. The code structure and execution encompass two main steps: the normalization of the Near-Infrared and Red bands of the multispectral images, and the construction of a three-dimensional Gaussian bell curve to visualize the distribution of vegetation indices using the meshgrid algorithmic technique. The results reveal a significant correlation between variations in the vegetation index and the vertical distribution of the Gaussian curve. Specifically, lower NDVI values indicate a diminished presence of vegetation or plant anomalies, resulting in an increase in the kurtosis of the Gaussian curve. To assess the effectiveness of the approach, Receiver Operating Characteristic analysis was employed, providing conclusive evidence regarding the reliability and performance of the implemented Python model.

Data mining is used in various fields, image processing is one of them, a particular application is the identification and classification of emotions expressed by students in the classroom. However, this creates challenges, such as the subjective interpretation of facial expressions and the need for extensive data sets to train and validate the models, for the former it is required to go to other allied research fields, and for the latter, a possibility is glimpsed in the transfer of learning. This work seeks to review and compare different classifiers for the construction of a model that allows the analysis of the emotions of university students from images extracted from recordings of face-to-face classes stored in an educational support platform. For this, the KDD (Knowledge Discovery in Databases) methodology was followed, and experiments were proposed with different configurations of hyperparameters and generation of models from classifiers such as Nearby Neighbors-KNN, Convolutional Neural Networks-CNN, and Random Forest. The performance of each one is contrasted based on precision, recall, F1, Accuracy, and ROC curve. Additionally, an approximation to a learning transfer process was carried out using an open-use data set (taken from the Kaggle repository) for the classification of emotions for the training of the models and validating with the data extracted from the source of the case study. The results support the utility and potential of applying these techniques in scenarios where image-based emotion analysis is required, with CNN being the classifier with the best accuracy and obtaining significant value from knowledge transfer that motivates further deepening of the approach for the treatment of this problem.

DevOps has significantly enhanced application operations through the utilization of containers and CI/CD. It still relies on human intervention in the event of failures in any system component. Many existing solutions are limited to specific issues, such as reacting to server outages and scaling them up. As the complexity of distributed systems continues to grow due to the simultaneous operation of numerous components, even minor unavailability can substantially impact application reliability and result in significant economic consequences for businesses. Therefore, it is imperative that the solutions being developed minimize risks and increasingly automate these operations. In light of these challenges, the emergence of AIOps offers a promising solution using artificial intelligence techniques, including machine learning and big data, to operate and maintain application infrastructures, reduce operational complexity, and automate IT operations processes. Implementing such solutions has been shown to improve system quality and significantly reduce the time it takes to detect errors and recover from them. These advancements mark significant progress in the realm of operations. However, despite these benefits, widespread adoption of AIOps solutions by most companies remains limited due to the challenges associated with implementing them in large projects and the lack of clear integration pathways for emerging solutions. In this paper, we propose a holistic architecture that facilitates the integration of cloud-native distributed systems with these new solutions.

Aedes aegypti, the vector responsible for transmitting diseases such as dengue, zika, and chikungunya, poses a significant public health threat in many regions. Understanding the dynamics of Aedes aegypti propagation is crucial for designing effective control and prevention strategies.Agent-Based Models (ABMs) have emerged as valuable tools for studying complex systems like vector-borne disease dynamics. Hybrid Agent-Based Models (HABMs), a variation of these models that incorporates Ordinary Differential Equations to model mosquitoes and ABMs to model humans, have been proposed by several authors.This study presents a comparative analysis of three HABMs to model Aedes aegypti propagation dynamics, with a focus on the impact of different modeling frameworks. The first model was built using Repast Simphony, a widely used ABM framework. It incorporates key factors such as mosquito life cycle, environmental conditions, and human-mosquito interactions. To enhance computational performance, the second model is migrated to a high-performance environment using Repast HPC. This migration leverages parallel computing capabilities to simulate larger populations. The third model is migrated to Mesa-Geo, a Python library specifically designed for geospatial agent-based modeling. This migration facilitates the integration of geospatial data into the model.Preliminary results show that migrating the model to a high performance environment enables more comprehensive analyses and reduces simulation runtime. Moreover, migrating to Mesa-Geo provides enhanced geospatial capabilities, and allows us to analyze the results in a graphical interface, which facilitates communication with decision makers.The main contributions of this research are: 1) insights into the trade-offs and benefits of using Repast Simphony, Repast HPC, and Mesa-Geo for modeling the transmission of viruses, and 2) a guide to researchers and stakeholders in selecting the most suitable modeling framework based on their specific requirements and available computational resources.

This article presents a predictive model to determine possible causes of commercial results in a company that commercializes products and services for the furniture and wood industry in Colombia. To achieve this, a literature review was carried out to identify analysis strategies and new technologies that could influence the proposed model. Then, using the CRISP-DM methodology, the main variables and indicators that make up the business model and the problems associated with decision-making were identified, with the aim of predicting and optimizing their KPIs, improving performance metrics and achieving an increase in commercial benefits. A case study was carried out with a data set of 99,972 records collected between 2020 and 2023, which facilitated the application of variable selection techniques to identify the most influential in the prediction. The model was developed using algorithms such as decision trees, random forests, and logistic regression. Once the model was trained, it was determined that the random forest regression algorithm with the Out-of-Bag validation method and an R2 of 94.1% provided the best results and delivered the highest sales prediction. In testing, the model showed that it was influenced by variables such as average invoice value, number of invoices, available inventory, and order fulfillment. These findings expand decision-making capacity by defining which variables must be controlled to improve results. In conclusion, The machine learning-based predictive model can identify potential causes of business outcomes and improve the accuracy of decisions at a strategic level in the area of timber trading. However, it is suggested to complement it with other variables to obtain an even more precise diagnosis.

This article describes the design and implementation of BDI agents for the WellProdSim Social Simulator, a system that assesses the productivity and social wellbeing of Peasant Families. A first BDI emotional reasoning model was designed to incorporate personal and social wellbeing components in the agent that represents a Peasant Family. Furthermore, decision-making mechanisms based on variable modulation and fuzzy logic evaluation of human welfare were added. The evaluation aspects include health state, knowledge and skills, food consumption, emotional state and expected productivity. Preliminary results demonstrate a high quality in the proposed model; although, some elements with potential for improvement, in future work, were also identified.

High employee turnover is a phenomenon that occurs in different types of companies and often leads to losses that could affect the organization’s productive continuity. This situation leads to the understanding, with solid evidence, of the factors that influence employees to leave their jobs and, thus, develop talent retention strategies proactively. This article proposes a predictive model to identify the most relevant factors that could cause employee turnover, specifically in the logistics process of a food and beverage production company. To achieve the objective, the CRISP-DM methodology was applied. Initially, various types of variables were identified, such as demographic, contractual, and payroll-related factors (N = 1517, period: 2017–2022). Then, five machine learning models, namely Logistic Regression, Random Forest, XGBoost, SVM, and AdaBoost, were trained, and optimal hyperparameters were used to improve the models’ performance and generalization. The performance evaluation of these models was conducted using classification metrics and the construction of confidence intervals for the accuracy metric through non-parametric Bootstrap. The results obtained demonstrate that the XGBoost and Random Forest models show the highest AUC value, with a result of 99%. This indicates that variables such as work environment, years of service, salary, workplace location, and monthly salary deductions are the most significant factors influencing the evaluated human talent to leave their job. Therefore, it is possible to conclude that the aforementioned two models are accurate and reliable for predicting employee turnover in the logistics process of the analyzed company.

Inclusion and accessibility are essential human rights that should be upheld in all aspects of life, including access to both public and private spaces. Unfortunately, in Colombia, the rights of people with disabilities are often marginalized and neglected, with only 7 out of 100 Colombians having disabilities in 2022. Furthermore, despite 36.9% of people with disabilities in Colombia living with reduced mobility, however, the cities remain largely inaccessible for them. This paper introduces Findby, a web application designed for users with reduced mobility, aiming to provide information on the accessibility of places and promote inclusive exploration through challenges and a reward system. The app utilizes user-centered design and leverages technology to improve access to both public and private spaces. Findby’s key features include accessibility markers, user-generated content, review comments, ratings, and customized route-based challenges, making the application engaging and user-friendly. Findby has the potential to contribute significantly to improving accessibility and inclusion for individuals with reduced mobility. Future work includes expanding the user community, improving the accuracy of accessibility information, incorporating accessibility information for private spaces, integrating additional features, and expanding globally to promote inclusivity and accessibility worldwide.

In traditional storytelling, a kind of predefined structure typically guides the narrative. However, when we move beyond the constraints of a rectangular screen, new possibilities emerge. 360-degree videos offer a unique opportunity to narrate with every element visible to the user in each frame. Nevertheless, some questions arise: What exactly are the end users seeing? Where is their attention directed? This paper deepens into an analysis of the data captured from the position and rotation of the headset worn by viewers while watching 360-degree videos. To accomplish this, two videos created by university students were examined, where gaze information was captured by testing it with a group of participants. Alongside watching the videos, participants provide their feedback on what elements they focused during the video. The findings revealed that participants tend to focus on elements intended to draw their attention. Furthermore, when the camera was stationary, participants found it easier to explore their surroundings, highlighting the value of utilizing the 360-degree format.

This article presents software that leverages virtual reality (VR) and OpenAI services to assess and detect early cognitive impairments, especially in verbal memory. The Thakira program, created for the Meta Quest 2, presents a natural environment with flying invertebrate animals and an interactive character. Through conversations simulated with artificial intelligence, the three-word recall test (R3P) is carried out on the user. We present the study design to evaluate the effectiveness of our method. This approach could help identify verbal memory impairment in people with neurocognitive disorders and alert them to the need for early professional attention.