Aprendizagem de máquina: modelo de predição de evasão acadêmica

Abstract

Dropout in higher education is a persistent challenge that impacts students and institutions. This work aims to apply data mining techniques to develop a predictive model based on machine learning capable of identifying patterns associated with academic dropout using socioeconomic variables. The research is characterized as a quantitative study, using historical information of students enrolled between the years 2012 and 2024 in a teaching degree program. The methodology involved rigorous preprocessing steps, encompassing the imputation of missing values, normalization, feature selection through statistical filtering, and class balancing using synthetic oversampling. Four supervised classification algorithms were trained and evaluated. The experiments demonstrated that information preservation associated with the elimination of irrelevant variables significantly enhanced the learning of the models. The algorithm based on random forests obtained the best performance and overall balance, presenting strong capacity for prediction, generalization and consistency in identifying students at risk. It is concluded that the adopted methodological approach proved technically viable, although the predictive performance obtained — still limited by the reduced volume of available data — indicates that the current results should be interpreted as a proof of concept rather than a model ready for direct application in institutional management, thereby constituting an initial step toward future decision-support tools aimed at student retention.