Conference Agenda

Student attrition in engineering leveling courses is a persistent challenge in Latin American higher education, often exacerbated by the lack of timely identification mechanisms. Traditional Early Warning Systems (EWS) frequently rely on mid-term grades, limiting their capacity for preventive intervention. This paper presents an Ensemble Learning model designed to predict final academic performance using exclusively pre-enrollment data, combining sociodemographic, previous academic records, and socioeconomic and psychometric variables. This study analyzed historical records of 1,533 students (7,210 instances) from a polytechnic university in Ecuador. Five machine learning algorithm families were evaluated through rigorous hyperparameter optimization involving 9,972 configurations. The proposed Stacking architecture, which integrates the top-performing XGBoost models via a Ridge Regression meta-learner, achieved a Coefficient of Determination (R2) of 0.749 and a Mean Absolute Error (MAE) of 3.783 on a 0-40 scale, demonstrating superior generalization compared to single models. Furthermore, SHAP (SHapley Additive exPlanations) analysis revealed that non-cognitive factors, such as study habits and academic self-confidence, are critical predictors alongside high school GPA and mathematics admission scores. To validate its practical utility, the model was deployed for the 2025-B cohort (N=900), identifying 9.2% of incoming students as "High Risk" before the start of classes. These results demonstrate the technical feasibility and ethical viability of using advanced ensemble methods for proactive, evidence-based academic management.

The present study addresses the limitations in the development of cognitive and academic skills of students in vulnerable educational contexts by evaluating an educational chess platform based on artificial intelligence. The objective was to analyze the impact of the interactive practice of this platform on secondary school students from a public institution in Cajamarca, Peru. A quantitative, applied approach and a pre-experimental design were used with a pre-test-posttest scheme, with a sample of 30 students. For data collection, the University Academic Performance Scale (RAU) was used, adapted to the secondary level, which presented adequate reliability with a Cronbach's alpha greater than 0.7. The results showed statistically significant improvements in the dimensions of objective academic performance, academic self-efficacy and commitment to learning, confirmed by the Student's t-test (p < 0.05). The results showed that the integration of chess with platforms based on adaptive artificial intelligence constitutes a viable pedagogical strategy to strengthen cognitive and academic skills in vulnerable educational contexts.

This research work develops an automatic system for the detection of suspicious attitudes through computational analysis of body postures, using the MediaPipe library to extract and process the anatomical landmarks of the human skeleton. The study starts from the fundamental premise that certain postural patterns, movement sequences, and biomechanical configurations can reveal potentially hostile intentions before they materialize into explicit actions, all with the aim of complementing current surveillance systems. The proposed system is trained using the MediaPipe algorithm for obtaining human skeletal landmarks in both two- and threedimensional spaces, considering three experimental scenarios: the detection of a hidden hand behind the body, the detection of a hand hidden under clothing at the front of the torso, and the recognition of suspicious gaze patterns. The system is evaluated using accuracy, precision, recall, and F1 score metrics. The results demonstrate high performance in all metrics and scenarios using three-dimensional spaces (95.6%, 95.7% and 97.3% respectively for accuracy metric). It suggests that the proposed system offers significant potential for applications in real security environments.

Urban mobility systems face increasing challenges due to congestion, unexpected traffic incidents, and ineffective control strategies that impact the sustainability and livability of cities. This study proposes an adaptive machine learning approach for real-time anomaly detection in vehicular data flows collected in a medium-sized Latin American city. The methodology integrates an Adaptive Random Forest (ARF) classifier for continuous learning on non-stationary data, addressing conceptual drift caused by dynamic traffic conditions. A public dataset of GPS vehicle trajectories and sensor readings was processed to identify anomalous patterns, such as congestion, accidents, or irregular flow behavior. Model performance was evaluated using confusion matrices, accuracy-recall analysis, and F1 score metrics, demonstrating robust adaptability to temporal variations in traffic density. The results highlight the potential of adaptive learning algorithms to improve sustainable traffic management, urban mobility planning, and decision support systems in smart cities by enabling early detection of anomalies and improving traffic efficiency

This study develops a data-driven framework to anticipate hemodialysis (HD) workload and erythropoietin (EPO) requirements in Paraguay’s public healthcare system using routinely collected service records from a national nephrology center (2021–2024; n = 2,592 patients). The objective is to support operational planning of critical resources by combining patient grouping strategies with simple univariate forecasting approaches. After systematic data cleaning and harmonization, temporal consistency adjustments were applied using institutional year-to-year statistics to preserve the continuity of the demand series. We compared empirical stratification with unsupervised grouping methods (PAM, hierarchical clustering, and DBSCAN), and evaluated several forecasting rules (naïve, Holt, linear regression, and weighted moving average). Given the short annual history, a single-year holdout (2024) was used to consistently screen configurations. Under this evaluation protocol, empirical stratification coupled with a weighted moving average achieved the lowest weighted mean absolute errors: 10.79 patients, 126.47 HD hours/week, and 116,938 IU/week of EPO. The selected configuration was then used to construct uncertainty bands for 2025, projecting 6,444–8,115 HD hours/week and 4.43–5.95 million IU/week of EPO to inform resource planning.

Learning analytics has become a strategic field in higher education, enabling the pedagogical interpretation of academic and contextual data for informed decision-making. This study aimed to evaluate the impact of using a virtual assistant based on generative artificial intelligence on the development of learning analytics competencies in postgraduate teacher training programs within the Master's Program in Digital Environments. A quasi-experimental design with a quantitative approach was employed, applying a pre-test and a post-test to 170 teacher trainees at the Bolivarian University of Ecuador. The results show statistically significant improvements in analytical skills, with a very large effect size in the statistical analysis dimension. Positive correlations were also identified between the use of the virtual assistant and its perceived pedagogical value. The findings confirm that the assistant acted as an effective pedagogical mediator, integrating statistical analysis and educational interpretation. It is concluded that the virtual assistant not only optimizes the analysis of educational data but also contributes to redefining learning analytics as a formative practice geared toward responsible and contextualized decision-making.

Keywords-- learning analytics, generative artificial intelligence, pedagogical decision-making; pedagogical mediation.