Predictive system for failure of basic education student

Abstract

Academic performance measures how students are doing according to the study plan proposed by schools. Educational performance is a crucial factor in education, making it important to develop a system that can indicate if a student is at risk of failing. This work discusses the development of a student failure prediction system for basic education using a dataset of about three million records of students' assessment grades from schools in Brazil. The system's primary objective is to identify students at risk of failing and provide indications to educators in search of preventing students' failure. The system architecture uses historical data from Middle and Basic Secondary Education students' assessments to predict approval or failure outcomes. Classification algorithms, including K-Nearest Neighbors, Decision Tree, Random Forest, and eXtreme Gradient Boosting, were applied for the system development, creating machine learning models. Results of model evaluations indicated that Random Forest and eXtreme Gradient Boosting showed the best performances. The final system utilizes eXtreme Gradient Boosting to create the predictive models due to its consistent performance and computational efficiency, achieving approximately 86% accuracy and F1-Score. These findings are implemented in a Business Intelligence Panel to empower educators to identify and assist students at risk of academic failure proactively.