Title:
Hybrid Machine Learning Pipeline for Predicting and Segmenting Academic Performance at a Mexican Public University [Download]Authors:
Rodolfo Alan Martínez Rodríguez, José Manuel Valencia-Moreno, Olivia Denisse Mejía Victoria, Alma Alejandra Soberano Serrano
Index Terms:
Abstract:
Academic performance prediction remains a critical challenge in Latin American higher education, where institutional heterogeneity and limited data availability constrain the effectiveness of traditional analytical approaches. While supervised machine learning models often achieve high predictive accuracy, they typically lack interpretability and fail to capture latent student profiles required for actionable educational interventions. This study proposes a unified and interpretable hybrid framework that integrates predictive modeling and clustering within a single analytical pipeline. The approach simultaneously estimates academic risk and identifies student profiles using a real-world dataset of 386 first-year students from a Mexican public university. The results demonstrate that, while individual supervised models achieved strong predictive performance (AUC up to 0.84), the primary contribution of the proposed approach lies in its integrative capacity. The hybrid pipeline achieved competitive performance (AUC = 0.85; F1-score = 0.76) and enabled cross-paradigm validation between predictive and clustering outputs ( $\kappa = 0.70$ ), providing evidence of structural consistency in student risk patterns. Beyond predictive accuracy, the framework links probabilistic predictions with behavioral profiles and anomaly detection, enabling the identification of heterogeneous at-risk groups and supporting targeted pedagogical interventions. These findings demonstrate the potential of hybrid learning analytics approaches to bridge the gap between predictive performance and actionable educational insights, offering a scalable and interpretable solution for early-warning systems in resource-constrained higher education environments.
DOI:
How to cite:
Rodolfo Alan Martínez Rodríguez, José Manuel Valencia-Moreno, Olivia Denisse Mejía Victoria, Alma Alejandra Soberano Serrano, "Hybrid Machine Learning Pipeline for Predicting and Segmenting Academic Performance at a Mexican Public University", IEEE-RITA, vol. 21, no. 1, pp. 319-330, Jan. 2026. doi: 10.1109/RITA.2026.3684879