Predictive System for Detection of Low Math Achievement in Peruvian Secondary Education Using Recurrent Neural Networks

Soto P.; Cutimbo J.; Burga D.; Tello C. · Proceedings of the 2025 IEEE 5th International Conference on Advanced Learning Technologies on Education and Research, ICALTER 2025 (2025) · DOI: 10.1109/icalter69698.2025.11355017

Resumen

Peru faces an educational emergency: in the latest PISA assessment, over 60 % of secondary students failed to reach baseline proficiency in mathematics. Traditional evaluation systems, focused solely on scores, lack predictive capabilities and overlook the multidimensional nature of student learning. This paper presents a predictive system based on Recurrent Neural Networks (RNNs), developed to detect early signs of low performance in mathematics. The model is trained using data structured under the PISA framework, integrating cognitive outcomes and subcompetencies such as quantity, change and relationships, and space and shape. It also incorporates socio-emotional, behavioral, and contextual variables to construct a comprehensive student profile. The system is a lightweight, teacher-facing web platform that surfaces calibrated risk; in prospective validation at the preregistered operating threshold, it achieved 0.91 of precision, enabling timely classroom action.

Instituciones: Universidad Peruana de Ciencias Aplicadas