Perspectives to Predict Dropout in University Students with Machine Learning

Abstract

This study analyzes the performance of four machine learning algorithms with different perspectives for defining data files, in the prediction of university student desertion. The algorithms used were: Random Forest, Neural Networks, Support Vector Machines and Logistic Regression. It was found that the Random Forest algorithm with 10 variables randomly sampled as candidates in each division, was the best for predicting dropouts and that the ideal perspective for training the algorithm is to use information on all semesters that students take within a given period of time, using a classification variable that defines the non-dropout as the graduated student. In a first validation sample, this approach correctly predicted 91% of dropouts, with a sensitivity of 87%.