Identifying Risky Dropout Student Profiles using Machine Learning Models

Abstract

Student dropout prediction is essential to measure the success of an educa- tion institute system. This paper focuses on identifying the dropout risk at the University of Évora based on student’s academic performance. Educa- tional data was collected from four different programs, from the academic years of 2006/2007 until 2018/2019. After gathering the raw data, some data pre-processing was done aiming to build a dataset capable of being used by Machine Learning algorithms. Decision trees, Naïve Bayes, Sup- port Vector Machines and Random Forests were evaluated, with the best model reaching an accuracy of around 96% when distinguishing between risky dropout and non-dropout students.