Predicting Elementary School GPA Using Machine Learning Approaches

Ivana Isailović

doi:10.19090/pp.v19i1.2625

Return

Predicting Elementary School GPA Using Machine Learning Approaches

2026-01-01

PRE

Ivana Isailović *

Isidora Gatarić

Ilija Milovanović

DOI：10.19090/pp.v19i1.2625

Original

Original request for help

Save

Abstract

En 中文

Predicting academic success, quantified as Grade Point Average (GPA), is one of the key research focuses in educational psychology, with ongoing debate regarding the most influential predictors. Previous studies suggest that cognitive readiness for school (e.g., attention, working memory, etc.) is among the strongest predictors of later academic achievement. In addition, social factors such as parents' education and gender also show consistent, though somewhat weaker, associations with GPA. However, many earlier studies have relied on traditional statistical models, such as linear regression, which assume linearity and often overlook complex, nonlinear interactions among variables. This limits their ability to uncover the true structure of influential predictors. In contrast, advanced machine learning (ML) methods, such as decision trees, random forest, and gradient boosting, can model such complexity, offering greater accuracy and deeper insight into predictor importance. This study applied these ML algorithms, along with linear regression, to predict GPA in 4(th )and 7(th )grade among 218 elementary school students, using measures of cognitive readiness and socio-demographic variables as predictors. Results indicated that linear regression and random forest yielded the most accurate predictions. The strongest predictors of GPA in both grades were measures of cognitive readiness (Coding, Visual Memory, General Knowledge, Block Assembly), while other predictors had minimal or no effect. These findings underscore the value of ML models in improving early identification of at-risk students and informing targeted academic support, while also illustrating the contexts in which traditional methods may still perform comparably.

Keywords:

elementary school GPA

decision trees

random forest

gradient boosting

cognitive readiness for school