Use of machine learning techniques to predict the graduation time of Computer Engineering students in the Southeast region of Brazil

Abstract

The National Student Performance Examination (ENADE) was created to assess student performance in courses at higher institutions. Through the performance of the students, it estimates the quality of the courses. Leaving or delaying the course leads to poor university management, since the budget that graduations receive is based on the number of graduating students. Analyzing ENADE data can generate insights into how long students remain in graduation. As the ENADE data contains a large amount of information, performing analysis visually is impracticable. To work around this situation, machine learning techniques can be used in order to automate this task and present the results. In this context, the objective of this work is to determine, through the ENADE 2019 database, the length of stay of students in graduation, considering Computer Engineering courses in the Southeast region of Brazil. The methodology involves pre-processing, feature selection, data balancing, Grid-Search parameter selection approach, cross-validation and classification. The results show that Random Forest performed well in the experiments carried out and that the application of SMOTE for data balancing is necessary.