Análise de fatores associados ao desempenho educacional no Brasil: uma abordagem por agrupamento com K-means nos dados da PeNSE e IDEB

Abstract

The 2022 PISA results revealed significant challenges for Brazilian education, with most students failing to reach minimum proficiency levels in mathematics, read ing, and science — a situation also reflected in IDEB indicators. Various factors, such as housing conditions, emotional and social aspects, and teacher training, contribute to this scenario. Despite the relevance of these assessments, there is still limited use of clustering techniques applied to national datasets to support educational policy development. This study aimed to identify patterns of student behavior, based on data from the National School Health Survey (PeNSE), that relate to different educational performance trajectories in IDEB. The K-Means clustering algorithm was applied to PeNSE microdata from 2009, 2015, and 2019 to identify behavioral patterns and associate them with educational outcomes. Students from the 9th grade of elementary school and the 3rd year of high school were analyzed, organized by federal unit and administrative sphere, using self-reported survey data. Preprocessing included cleaning inconsistencies, imputing missing values, and standardizing variables. The number of clusters was determined using metrics such as elbow, silhouette, Calin ski–Harabasz, and Davies–Bouldin. Between three and four clusters were identified per edition, and the results showed that low parental supervision, poor diet, sedentary behavior, and early exposure to alcohol and drugs were associated with lower IDEB performance, especially in public schools. Despite limitations — such as the heterogeneity of PeNSE cycles and the absence of a direct link between behavior and performance — the study reveals recurring patterns and provides valuable insights for more effective public policies. Factors such as parental supervision, physical activity, and healthy eating emerge as priorities for future interventions.