Análise de evasão baseada em modelos preditivos para o curso de Engenharia de Computação da Universidade Federal de Alagoas

Abstract

Context: School dropout is a critical and chronic problem present at different levels of teaching in Brazilian education. This problem is extremely present in public colleges, especially in exact sciences courses, directly affecting the Computer Engineering course at the Institute of Computing (IC) of the Federal University of Alagoas (Ufal). Objective: Cthere is no resource being used to help the coordination of the referred course to map students with high dropout risk. Therefore, the objective of this work is to evaluate the feasibility of using tools based on predictive models as a means of support for reducing school dropout rates. Method: This work presents the comparison between two predictive models, the first based on simple decision trees and the second based on the Random Forest algorithm using the database of students of the Computer Engineering course at IC/Ufal. These data were provided by the Information Technology Center (NTI) of the institution and contain information from the beginning of the course until the first half of 2020. Results: The results indicate that the generated models have accuracy and recall above 85% in both models, results considered positive given the research problem presented. In addition, other metrics and relevant information could be extracted from the data used. Conclusion: In view of the work carried out and the results obtained, we believe that the use of predictive models based on decision trees as a tool to support the dropout reduction in the Computer Engineering course at IC/Ufal is feasible.