Educational inequalities in enem: a perspective based on socioeconomic variables and machine learning

Abstract

The National High School Exam serves as an important gateway to higher education in Brazil. This study examines the relationship between socioeconomic variables and student performance on the exam, employing machine learning techniques to identify significant patterns. The research has three main objectives to develop predictive models based on random forest to classify student performance to identify the most relevant socioeconomic variables and to analyze their impact on results, aiming to inform more equitable educational policies. The study used Enem 2023 microdata, which underwent preprocessing including one-hot encoding for certain variables and Smote for balancing. Ten random forest models were built, with hyperparameter tuning via random search. Performance was evaluated using metrics such as accuracy, precision, recall, and F1-score, along with variable importance analysis. The models demonstrated satisfactory performance, with accuracy around 94% and precision up to 99%. Parental education level, occupation, and family income emerged as key predictors. Students with more educated parents in strategic professions were three times more likely to achieve high performance, while those from low-income families showed greater tendency toward unsatisfactory results. The findings highlight the influence of socioeconomic factors on educational performance, underscoring the need for appropriate public policies. The models' effectiveness confirms their utility for educational diagnostics.