Banca de QUALIFICAÇÃO: ANDERSON KLEY B CAMARA PINHEIRO

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
STUDENT : ANDERSON KLEY B CAMARA PINHEIRO
DATE: 16/04/2026
TIME: 09:00
LOCAL: https://meet.google.com/pce-jjtq-wtm
TITLE:

 

Predicting Educational Risk in Rio Grande do Norte:

Application of Machine Learning and XAI to SAEB microdata.

KEY WORDS:

GovTech; public policies; school dropout; predictive algorithms; social inequalities.

PAGES: 101
BIG AREA: Ciências Sociais Aplicadas
AREA: Desenho Industrial
SUMMARY:

The landscape of public education in Rio Grande do Norte faces chronic challenges related to school failure, historically mitigated through a strictly reactive and retrospective management approach. Situated at the intersection of information technology and public administration, this research aims to develop and validate a predictive model based on artificial intelligence to anticipate underlying factors of student vulnerability, enabling preventive interventions through the Jogabilizar platform. The methodology adopted was based on the CRISPR-DM model, using official microdata from the Basic Education Assessment System (SAEB) for the 2023 cycle. Pre-processing included exclusive filtering of students from the public school system, cleaning of missing data, and risk assessment based on the calculation of the internal proficiency average. For modeling, the XGBoost algorithm was trained, which exhibited highly conservative and protective mathematical behavior, achieving an Overall Accuracy of 78%, an Area Under the ROC Curve (AUC-ROC) of 0.865, and a Recall of 70% in identifying students at risk of failure. In order to overcome algorithmic opacity and meet the demands of governmental transparency, Explainable Artificial Intelligence was applied using SHAP values. Game theory revealed that the main drivers of risk are not limited to the classroom, being strongly anchored in structural socioeconomic vulnerability (lack of consumer goods), the absence of time dedicated to extracurricular study, a history of previous grade repetition, high family cohabitation, and students' perception of decontextualized pedagogical practices. It is concluded that the integration of big data with explainable algorithms provides public management with a surgical diagnostic tool. By operating under the premise of human intervention supported by data, the solution consolidates the transition from a reactive government model to the design of evidence-based preventive public policies

COMMITTEE MEMBERS:
Presidente - ***.133.498-** - CRISTIANO ALVES DA SILVA - UFSC
Interno - 1943220 - ORIVALDO VIEIRA DE SANTANA JUNIOR
Interna - 1753896 - ZULMARA VIRGINIA DE CARVALHO
Externa à Instituição - MARILIA MATOS GONÇALVES - UFSC