Previsão de evasão estudantil na Faculdade de Computação utilizando mineração de dados

Abstract

Student dropout in higher education is a growing problem in Brazil. According to the Higher Education Map, published by Semesp in 2024, around 57.2% of students did not complete their undergraduate degrees between 2018 and 2022. In this context, this study employs data mining, machine learning, and statistical analysis techniques to uncover patterns and relevant information from a real dataset of students enrolled in the Bachelor’s Degree in Computer Science (BCC), provided by the coordination of the Computer Science course at the Faculty of Computing of the Federal University of Uberlândia (FACOMUFU). The dataset spans a period of 27 years and includes the academic records of approximately 1,500 students, containing information such as completed courses, grades, completed credit hours, among other indicators. Based on academic data from the first three months of the course, balancing techniques were applied in combination with several supervised learning algorithms to develop models capable of identifying students at higher risk of dropping out of higher education. The main goal is to evaluate the performance of these models so they may eventually support the institution in decision-making aimed at promoting the well-being of students in situations of academic vulnerability. The results show that the algorithms generally performed well, with stable identification of classes within the dataset, although they showed limited sensitivity to the applied balancing techniques. Random Forest, Naive Bayes, and MLP models stood out, albeit with only minor differences among them.