Prediction of academic situation in undergraduate students using machine learning algorithms.

Abstract

The academic performance of a university student is generally measured through their grades, which result in a normal or poor interpretation of the academic performance of the students. The grades, actually depends on various factors. The objective of this research was to find the main predictors of the academic performance of a university student after six semesters since her or his admission. For data analysis, the Boruta algorithm was used to select predictor variables and twelve classification algorithms were applied, after partitioning the data into training and evaluation sets. Then, those models with the best sensitivity, specificity and balanced accuracy values were chosen. Finally, an optimal assembly and cut-off point were used to improve predictions. The models with the best performance were logistic regression, Naive Bayes and vector support machines with linear kernel. The application used ensembles with optimal cut-off point, specificity of 0.695 and sensitivity of 0.947 were obtained. The grade obtained in Mathematics course was one of the most important to predict the academic performance after six semesters of studies, while the sociodemographic variables were not relevant.