Development of a Student Performance Prediction System with Deep Learning

Villegas-Espinoza A.E.; Necochea-Chamorro J.I. · TEM Journal (2025) · DOI: 10.18421/tem144-64

Resumen

The present study focuses on the development of deep learning (DL) model to predict student performance and address the issue of low academic achievement among secondary school students. Utilizing a quasi-experimental design, the research process involved the enhancement of data with generative adversarial networks (GAN) and synthetic minority over-sampling techniques (SMOTE), followed by the training of a deep neural network (DNN). The model demonstrates high accuracy, achieving 96.81% and a Cohen's Kappa index of 0.866, indicating strong reliability. A comprehensive investigation is conducted to identify the key variables influencing student performance. These variables include self-concept, attitude towards subjects, parental satisfaction, and the use of additional learning resources. These factors are critical in building a robust predictive model capable of detecting students at risk of poor academic outcomes. The findings underscore the efficacy of the model in not only identifying at-risk students but also in facilitating the implementation of personalized intervention strategies. The objective of these strategies is twofold: to reduce the rate of school failure and to enhance the overall quality of education.

Instituciones: Universidad César Vallejo