Prevendo o desempenho escolar usando machine learning e habilidades socioemocionais

Abstract

There is a diversity of studies that show non-cognitive skills as important for the development of students in school. However, these studies do not have an approach that seeks to study the interactions between these skills and the non-linearity present between them. In this work, the SENNA instrument was used, built to measure social and emotional skills in Brazilian schools. The instrument was applied by the Ayrton Senna Institute to 105,594 students in Ceará in 2015. Using machine learning methods, an attempt was made to capture non-linear effects and predict school performance in Portuguese and mathematics through socio-emotional skills. It was noticed that the Random Forest model was 16% better than the linear model in predicting student performance and that the addition of interactive variables improves the prediction model. In predicting the worst students in the Portuguese subject, the Random Forest model had an accuracy of 76.4%. It was possible to conclude that machine learning models have the property to predict student performance and there is also room to use other newer machine learning methods that use more computational power.