Predictive Models using Artificial Intelligence: Early Detection of Communicative Diversity Profiles in Higher Medical Education

Flor García Calle D.; Lorena Becilla Vera M.; Isabel Cordero Alvarado N.; Graciela Urgilés Carabajo C.; Mariela Solano Farias K.; Asunción Ortiz Salas D.; Jacqueline Mendoza Vega A. · Revista Cubana de Educacion Medica Superior (2026)

Resumen

Introduction: Early detection of communicative diversity profiles represents a contemporary line of research that can support effective educational processes in the health sciences through the integration of artificial intelligence. Objective: To establish an artificial intelligence-supported predictive model for the early identification of communicative diversity profiles in students of higher medical education. Methods: A quantitative, applied study with an explanatory-predictive scope and a non-experimental, cross-sectional design was conducted. A total of 200 health sciences students participated and were assessed using communicative, cognitive, and metacognitive instruments. Data were analyzed using supervised artificial intelligence models (logistic regression, SVM, Random Forest, Gradient Boosting, and neural networks), employing stratified cross-validation and predictive performance metrics. Results: Early detection of communicative diversity profiles showed the highest predictive performance with ensemble models, with Gradient Boosting standing out (AUC-ROC = 0.90±0.02; F1-score = 0.87; sensitivity = 89.1%). Random Forest achieved statistically comparable performance (AUC-ROC = 0.88±0.02), with greater stability across folds (CV < 4%). Significant differences were observed between logistic regression and Gradient Boosting (ΔAUC = 0.10; p<0.001), but not in comparison with Random Forest (p>0.05). Interpretability analysis identified reading comprehension (24.7%), academic oral communication (28.4%), and self-regulated learning (19.6%) as the main predictive variables. Inter-rater agreement among expert evaluators was high (Kappa = 0.79). Conclusions: Ensemble models exhibit substantially higher predictive performance for the early detection of communicative diversity profiles.

Instituciones: Universidad César Vallejo