Using machine learning techniques to predict academic performance in mathematics

• Published in: EDUWEB Vol. 19; no. 2; pp. 9 - 19
• Main Authors: Batista da Silva, João, Assis da Costa, Luis Carlos, dos Santos, José Nilson, da Silva Batista, Raquel Oliveira
• Format: Journal Article
• Language: English; Spanish
• Published: 30.06.2025
• ISSN: 1856-7576; 2665-0223; 1856-7576
• DOI: 10.46502/issn.1856-7576/2025.19.02.1

The purpose of this study was to investigate the predictive power of the SAEF exams in estimating the schools' performance in Mathematics on the SPAECE exam. To achieve this, we developed a predictive machine learning model. The model was trained using data from 133 schools that participated in the exams in 2022, and subsequently tested with data from 140 schools that took part in the exams in 2023. The results showed that the random forest (RF) model demonstrated moderate predictive power (R² = 0.397), which was superior to the linear model (R² = 0.384). This means that approximately 39.7% of the variance in schools' Mathematics performance on the SPAECE can be explained by the results of the SAEF exams. The first SAEF exam, administered at the beginning of the academic year, demonstrated the highest predictive power among the three, indicating that students' initial performance in Mathematics is a strong indicator of their future performance. These findings underscore the importance of early identification of learning difficulties to enable strategic pedagogical interventions throughout the year. Although this study was conducted within the Brazilian educational context, other countries can also utilize machine learning techniques to monitor students' academic trajectories and predict their outcomes in standardized assessments.