Resumo

POZO VALDIVIEZO, Eduardo; TENELANDA CUDCO, Sandra Elizabeth; DAVALOS VILLEGAS, Martha Ximena  e  AVILA GAIBOR, Gustavo Javier. Explainable Models for Identifying Difficulties in University Mathematics: Analysis of the MathE Dataset. Athenea [online]. 2026, vol.7, n.23, pp.66-76.  Epub 11-Abr-2026. ISSN 2737-6419.  https://doi.org/10.47460/athenea.v7i23.134.

This study applies a learning analytics approach and explainable models to identify the most difficult content areas in university mathematics based on interaction records from a practice and assessment platform. A secondary quantitative analysis of the MathE dataset was conducted, considering content variables (topic, subtopic, and keywords) and contextual variables (country and level). First, error rates were estimated by topic and subtopic, and patterns were synthesized through comparative visualizations. Then, complementary models were trained: a logistic regression model for its interpretability and a nonlinear gradient-boosted tree model to capture interactions, validating generalization through student-level partitioning. Explainability was addressed through contribution attribution to interpret factors associated with errors. The findings indicate greater difficulties in Differentiation, Functional Interpretation, and Probability, together with cross-cutting weaknesses in algebraic manipulation, with additional support needs in Numerical Methods and Integration.

Palavras-chave : learning analytics; mathematics education; explainable artificial intelligence; instructional guidance.

        · resumo em Espanhol     · texto em Espanhol     · Espanhol ( pdf )