Resumen

RICO PAEZ, Andrés  y  GAYTAN RAMIREZ, Nora Diana. School exercises for predicting university students' academic performance using combined machine learning techniques. RIDE. Rev. Iberoam. Investig. Desarro. Educ [online]. 2026, vol.16, n.32, e1041.  Epub 30-Abr-2026. ISSN 2007-7467.  https://doi.org/10.23913/ride.v16i32.2835.

The purpose of this study is to propose a methodology for developing predictive models of student academic performance using academic exercises completed in class and combined machine learning techniques known as majority voting and stacking. Data were collected from 250 university students in Mexico regarding their assessments of school exercises to develop the models, and performance metrics were obtained through cross-validation. Subsequently, the constructed models were applied to 108 students in a later semester of the same course, and their metrics were calculated.

The results obtained through cross-validation show that the stacking technique with the k-nearest neighbors’ method in the second phase has the highest accuracy (69.2%). When predicting the academic performance of 108 students using the developed models, the highest accuracy is obtained with the stacking technique that includes the k-nearest neighbors’ method in the second phase, with a value of 74.1%. The information obtained was collected 17% of the way through the course, facilitating the early detection of students with academic difficulties so that teachers can intervene promptly and improve their performance. It is common for teachers to collect assessments of academic exercises without needing to use more complex data collection tools, which favors the use of this type of methodology for building predictive models.

Palabras llave : combined machine learning techniques; stacking; majority voting; predictive models; higher education.

        · resumen en Español | Portugués     · texto en Español     · Español ( pdf )