Investigating Student Dropout Risk in Higher Education through Machine Learning

Abstract

In recent years, there has been a significant increase in the student dropout rate in Higher Education. Various reasons are cited, such as difficulty learning the content, the proposed course structure, and lack of financial resources. This study explores machine learning (ML) in the student dropout problem. The experiments were conducted with a dataset from the Academic System of a Federal University listing different academic features of the students. They showed promising results, with Random Forest accurately predicting the student situation with an average F-Score of 0.959. However, most relevant features are expected and do not provide any new insight regarding the dropout imminence. Future experiments can fix it with a more robust feature engineering process.