DIAGNOSTIC ANALYSIS OF MATHEMATICS LEARNING PROFILES IN SAEB: CONTRIBUTIONS OF EXPLAINABLE ARTIFICIAL INTELLIGENCE AND HISTORICAL-CULTURAL MEDIATION IN SECONDARY EDUCATION

Source abstract

This article presents the development, application, and analysis of SAB-IA+, a system for diagnostic analysis and interpretation of learning profiles designed for mock exams in the format of the Basic Education Assessment System (SAEB) in Mathematics, in the third year of secondary education. The system links three computational components, a diagnostic proficiency estimation model based on the LightGBM algorithm, an explainability method based on SHAP (SHapley Additive exPlanations), and a student clustering algorithm based on response-pattern similarity using K-Means, to the principles of Historical-Cultural Theory, drawing on the contributions of Vygotsky, Luria, Leontiev, and Davydov, in dialogue with critical references of Brazilian education, such as Freire, Hoffmann, and Luckesi. The research is configured as a single case study, qualitative and exploratory in nature, with characteristics close to action research, conducted in a state public school in Santa Catarina over three cycles of mock exams applied between July and September 2025, involving 71 unique third-year secondary students. The results indicate that the system expanded the level of diagnostic detail available to the teacher, supporting the organization of pedagogical interventions differentiated by learning profiles, without replacing the centrality of teacher mediation. The technical, ethical, and generalization limits of the study are also discussed, as well as perspectives for future large-scale investigations.